The evolution of 21st century sea-level projections from IPCC AR5 to AR6 and beyond

Abstract

Sea-level science has seen many recent developments in observations and modelling of the different contributions and the total mean sea-level change. In this overview, we discuss (1) the evolution of the Intergovernmental Panel on Climate Change (IPCC) projections, (2) how the projections compare to observations and (3) the outlook for further improving projections. We start by discussing how the model projections of 21st century sea-level change have changed from the IPCC AR5 report (2013) to SROCC (2019) and AR6 (2021), highlighting similarities and differences in the methodologies and comparing the global mean and regional projections. This shows that there is good agreement in the median values, but also highlights some differences. In addition, we discuss how the different reports included high-end projections. We then show how the AR5 projections (from 2007 onwards) compare against the observations and find that they are highly consistent with each other. Finally, we discuss how to further improve sea-level projections using high-resolution ocean modelling and recent vertical land motion estimates.

Impact statement

Sea-level rise is an important aspect of climate change, with potentially large consequences for coastal communities around the world. Sea-level change is therefore an active area of research that has seen many developments in the past decades. Based on the available research, the Intergovernmental Panel on Climate Change (IPCC) provides regular updates on sea-level projections which are used by policymakers and for adaptation planning. In this review, we compare the sea-level projections from different IPCC reports in the past 10 years and explain what has changed in the methods used and in the numbers presented. We also compare observed changes from the 2021 IPCC report to projected changes from the 2013 IPCC report, for the overlapping period 2007–2018, and find that they are highly consistent. Finally, we share some potential future research directions on improving sea-level projections.

Introduction

Present-day sea-level change (SLC) is primarily a consequence of human-induced climate change, which will impact people and communities all over the world. From a decision-making perspective, knowing how much the sea level will rise, and when, can help to decide which protective measures need to be taken at which point in time. Therefore, sea-level projections are among the most anticipated outcomes of the Intergovernmental Panel on Climate Change (IPCC) assessment reports (ARs). While sea-level extremes are also an important consideration for future coastal hazards, in this review we focus our attention on projections of mean sea level.

In the past 15 years, process-based sea-level projections (i.e., projections which use models to simulate the physical processes and interactions contributing to sea-level change) in the IPCC reports have developed from global-mean only (AR4, Meehl et al., 2007) to regional projections (AR5, Church et al., 2013a). More recent reports focused on the Antarctic contribution (Special Report on Oceans and Cryosphere in a Changing Climate SROCC, Oppenheimer et al., 2019), and provided projections consistent with the assessed Equilibrium Climate Sensitivity (ECS) (AR6, Fox-Kemper et al., 2021a). The research community has dedicated significant research effort and published many papers on improving the understanding and modelling of the different contributions to SLC, such as ice sheets, glaciers and sterodynamic changes (e.g., Gregory et al., 2016; Nowicki et al., 2016; The IMBIE Team, 2018, 2019; Hock et al., 2019). Since AR5, new global mean and regional projections have been published, using various methods: for instance based on fully coupled climate models (Slangen et al., 2012; Kopp et al., 2014; Slangen et al., 2014a; Carson et al., 2015; Jackson and Jevrejeva, 2016; Buchanan et al., 2017; Palmer et al., 2020), reduced-complexity models (Perrette et al., 2013; Schleussner et al., 2016; Nauels et al., 2017), semi-empirical models (Kopp et al., 2016; Mengel et al., 2016; Bakker et al., 2017; Bittermann et al., 2017; Goodwin et al., 2017; Wong et al., 2017; Jackson et al., 2018; Jevrejeva et al., 2018), structured expert judgement (SEJ) (Bamber et al., 2019), or a mixture of methods (Grinsted et al., 2015; Kopp et al., 2017; Le Bars et al., 2017; Le Cozannet et al., 2017a). There have also been a number of reviews, including a database of sea-level projections (Garner et al., 2018), reviews on developments following AR5 (Clark et al., 2015; Slangen et al., 2017a), overviews of processes and timescales (Horton et al., 2018a; Hamlington et al., 2020), reviews on coastal sea-level change (e.g., Van de Wal et al., 2019) and reviews integrating risk and adaptation assessments (e.g., Nicholls et al., 2021).

One thing that all sea-level projections have in common, despite the different approaches and methodologies, is an uncertainty that grows substantially through time. The uncertainties in regional sea-level projections over the coming years to decades result primarily from internal climate variability (see e.g., Palmer et al., 2020, their Figure 11). On decadal to centennial timescales, uncertainties depend on the future forcings (such as greenhouse gas emissions) and the response of the climate system; and on the modelling uncertainty associated with simulating the different contributions to SLC. The forcing uncertainty can be assessed using different emissions or radiative forcing scenarios, varying from scenarios with net-zero CO₂ emissions by 2050 to scenarios with a tripling of the present-day CO₂ emissions by 2100. The modelling uncertainty can be relatively well quantified for some contributions, such as global mean thermal expansion. For other contributions, such as (multi)-century timescale ice mass loss of the Antarctic Ice Sheet, the uncertainty is characterised as ‘deep uncertainty’, which means that experts do not know or cannot agree on appropriate conceptual models or the probability distributions used (Lempert et al., 2003; Kopp et al., 2022). These contributions are therefore a topic of much research and debate (e.g., Oppenheimer et al., 2019).

In addition to studies on future sea-level projections, much research has been focused on understanding past observations. A lot of progress has been made in the closing of the sea-level budget for the 20th century, which compares the sum of the observed contributions to the total observed changes, on global (e.g., Gregory et al., 2013; Chambers et al., 2016; Cazenave et al., 2018; Frederikse et al., 2020) and basin scales (e.g., Slangen et al., 2014b; Frederikse et al., 2016; Rietbroek et al., 2016; Frederikse et al., 2018; Wang et al., 2021b). These budget studies have led to important advances in the understanding of sea-level change and its contributing processes on global and regional scales. In addition, the observations can be compared with model simulations (Church et al., 2013c; Meyssignac et al., 2017; Slangen et al., 2017b; Oppenheimer et al., 2019) to test, understand and improve the model representation of the different processes. This has turned out to be challenging, especially for the earlier part of the 20th century: SROCC stated that only 51% of the 1901–1990 observed global mean sea-level (GMSL) change could be explained by models, due to ‘the inability of climate models to reproduce some observed regional changes’, in particular before 1970. The agreement between models and observations increased to 91% for 1971–2015 and 99% for 2006–2015 (Oppenheimer et al., 2019).

It is also possible to evaluate past projections against observations that have been made since. For instance, for total SLC, Wang et al. (2021a) found an almost identical GMSL trend in the observations and AR5 projections for the period 2007–2018. Lyu et al. (2021) compared observations and climate model output of ocean warming for the purpose of constraining projections. They found a high correlation for the Argo period (2005–2019) and concluded that the observational record over this period is currently the most useful constraint for projections of ocean warming. Such evaluations against the already realised SLC are important to provide further insights and build confidence in sea-level projections.

Here, we will first discuss ‘how we got here’: recent methodological developments in process-based sea-level projections for the 21st century, with a brief recap of the IPCC sea-level projection methods up to IPCC AR5, followed by a discussion of the key differences between AR5, SROCC and AR6 projections (section ‘Key advances in sea-level projections up to IPCC AR6’). Next, we discuss ‘where we are’, by evaluating AR5 projections of SLC (which start from 2007 onwards) against observational time series (up to 2018), both for total GMSL and for individual contributions (section ‘Comparison of the AR5 model simulations with observations’). Finally, we discuss ‘where we’re going’: how can sea-level projections be better tailored for coastal information (section ‘Moving towards local information’). Throughout this review, we adopt the sea-level terminology defined by Gregory et al. (2019) and we refer to Box 9.1 of IPCC AR6 (Fox-Kemper et al., 2021a) for a summary of the key drivers of SLC.

Key advances in sea-level projections up to IPCC AR6

There have been substantial methodological and scientific advances in sea-level projections since the publication of the IPCC First Assessment Report in 1990 (Warrick and Oerlemans, 1990). The use of global climate models (GCMs) in IPCC sea-level projections dates back to the IPCC Third Assessment Report (Church et al., 2001). In IPCC AR4, climate models from the third phase of the Climate Model Intercomparison Project (CMIP3) were used as the ‘backbone’ of the process-based GMSL projections (Meehl et al., 2007), with a similar approach adopted for AR5 using the CMIP5 generation of climate models (Church et al., 2013a). A major change for the AR5 was the inclusion of regional projections (following Slangen et al., 2012). The IPCC Special Report on Global Warming of 1.5° for the first time assessed GMSL based on warming levels (Hoegh-Guldberg et al., 2018). The SROCC added new information on the dynamical ice sheet contribution to the AR5 projections (Oppenheimer et al., 2019). The main advance in AR6 was the use of physics-based emulators to ensure consistency of the sea-level projections with the AR6-assessed ECS and global surface air temperature (GSAT).

We will now discuss some of the key differences of the global mean and regional sea-level projections in IPCC AR6 relative to AR5 and SROCC, by explaining what has been done differently, why these changes were made, and what the effects are on the projections. We do not include the SR1.5 projections in this discussion (Hoegh-Guldberg et al., 2018), as the SR1.5 report made a literature-based assessment of GMSL changes for 1.5° and 2°, but did not produce new projections.

Before we discuss the projections, we note that the interpretation and communication of the uncertainties in sea-level projections has varied across the different IPCC assessment reports (Kopp et al., 2022). IPCC reports use calibrated uncertainty language, in which the confidence level is a qualitative reflection of the evidence and agreement, whereas the likelihood metric is a quantitative measure of uncertainty, expressed probabilistically (Box 1.1, Chen et al., 2021). For the medium confidence projections in AR5, the 5–95th percentile range of the model ensemble was interpreted as the likely range (the central range with about two-thirds probability, 17–83%), with the uncertainty range of all contributions inflated relative to the model spread to account for structural uncertainties arising from the CMIP5 model ensemble. For the sea-level projections in AR6 (Fox-Kemper et al., 2021a, Section 9.6), the likely range was redefined as the central range with at least two-thirds probability, encompassing the outer 17th to 83rd percentiles of the probability distributions considered in a p-box (e.g., Le Cozannet et al., 2017a). That is, the definition of likely range in AR5, SROCC and AR6 is comparable but not exactly the same, and the way of determining the range from the available information is different. The AR6 medium confidence projections include estimated distributions for each emissions scenario with two different methodological choices for the Antarctic ice sheet (see Table 1). AR6 also presented a set of lowconfidence projections, which include additional contributions from ice sheet processes and estimates for which there is less agreement and/or less evidence (see Table 1).

Table 1. High-level summary of the methods used in the AR5, SROCC and AR6 reports to project global mean and regional SLC (1° × 1° resolution) to 2100

Note. This is an adapted version of Table 9.7 in Fox-Kemper et al. (2021a).

The methodologies of the projections in AR5, SROCC and AR6 are briefly summarised in Table 1; for more details, we refer to Chapter 13 of AR5 (Church et al., 2013a), Chapter 4 of SROCC (Oppenheimer et al., 2019) and Chapter 9 of AR6 (Fox-Kemper et al., 2021a). We focus on three major elements of the projections that have changed: (1) the use of CMIP5 versus CMIP6 model output and consistency with the assessed ECS (section ‘Updated climate model information and the use of emulators’); (2) differences in the approaches to project the contributions to SLC (section ‘Differences in the projected contributions to SLC’); (3) the way the reports addressed potential outcomes outside the likely range (section ‘Sea-level projections outside the likely range’).

Updated climate model information and the use of emulators

The majority of the sea-level projections for the 21st century since AR5 have been based on CMIP5 climate model output (Taylor et al., 2012), forced by Representative Concentration Pathways (RCP, Meinshausen et al., 2011), which are scenarios of future greenhouse gas concentrations and aerosol emissions. The projections in AR6 used information from CMIP6 climate models (Eyring et al., 2016), which were forced by Shared Socioeconomic Pathways (SSP, O’Neill et al., 2014): scenarios of socio-economic development (including for instance population change, urbanisation and technological development) in combination with radiative forcing changes (GHG emissions and concentrations). These scenarios are noted as SSPx − y, where x denotes the SSP pathway (SSP1 sustainability, SSP2 middle-of-the-road, SSP3 regional rivalry, SSP4 inequality, SSP5 fossil fuel-intensive) and y the radiative concentration level in 2100 in W/m². AR6 used five illustrative SSP scenarios: SSP1–1.9 (very low emissions), SSP1–2.6 (low emissions), SSP2–4.5 (intermediate emissions), SSP3–7.0 (high emissions) and SSP5–8.5 (very high emissions).

The ECS from the CMIP6 model ensemble has a higher average and a wider range compared to the CMIP5 model ensemble and compared to the AR6 assessment of ECS (Forster et al., 2021). The consequences of this change in ECS distribution for projections of GMSL change were investigated by Hermans et al. (2021), who used CMIP6 data in combination with the AR5 methodology. They found that, while the projected change in GSAT median and range increased substantially from CMIP5 to CMIP6 (from 1.9 (1.1–2.6) K to 2.5 (1.6–3.5) K under SSP2-RCP4.5, see their Table S2 for additional scenarios), the upper end of the GMSL likely range projections at 2100 increased by only 3–7 cm across all scenarios (see their Figures 1 and 3), due to the delayed response of SLC to temperature changes. However, they also found an increase in the end-of-century GMSL rates of up to ~20%, suggesting that differences between CMIP5 and CMIP6-based GMSL projections could become substantially larger on longer time scales.

Figure 1. Comparison of 21st century projections of global mean SLC in AR5, SROCC and AR6. Total GMSL and individual contributions, between 1995 and 2014 and 2100 (m), median values and likely ranges of medium confidence projections, for (a) RCP2.6/SSP1–2.6 and (b) RCP8.5/SSP5–8.5. See also Table 9.8 in Fox-Kemper et al. (2021a) for comparative numbers of GMSL projections. AR6 low confidence projections for SSP1–2.6 and SSP5–8.5 in grey for Greenland, Antarctica and GMSL. Corrections for the differences in baseline period between AR5 (1986–2005) and AR6 (1995–2014) were done following IPCC AR6, Table 9.8.

Figure 2. Comparison of regional relative sea-level change w.r.t. the global mean sea-level change in AR5 and AR6 (2020–2100) (%), based on median values, for (a) IPCC AR5 RCP4.5, global mean of 0.46 m and (b) IPCC AR6 SSP2–4.5, global mean of 0.51 m.

Figure 3. Comparison of observations (IPCC AR6, available up to 2018) and projections (IPCC AR5, available from 2007) of GMSL change. (a) Total GMSL and (b-f) individual contributions in (m) with respect to the period 1986–2005; all uncertainties recomputed to represent the likely range. Text in panels compares rates (mm/yr) of observations for 2006–2018 (Fox-Kemper et al., 2021a, Table 9.5) to rates of projections for 2007–2018 (Church et al., 2013a); rates rounded to nearest 0.1 mm/yr; time periods used for rates differ by 1 year, allowing for traceability to the IPCC reports. Note that AR5 included the Greenland peripheral glaciers in the glacier contribution, whereas AR6 included it in the Greenland contribution; we have therefore subtracted a Greenland peripheral glacier estimate of 0.1 mm/yr from the AR6 Greenland observations in (c) and added it to the AR6 glacier observations in (d), both for the time series and the rates (Church et al., 2013a, Table 13.1; Fox-Kemper et al., 2021b, Table 9.SM.2). The AR5 observed glacier change is added to (d) for reference (using the 1993–2010 linear rate from Table 13.1 of Church et al., 2013a).

One of the novel aspects of AR6 was the use of a physically-based emulator, which allowed for projections of 21st century GSAT and SLC that were consistent with the AR6 assessment of ECS (Forster et al., 2021). The AR6 used a simple two-layer energy balance model (e.g., Geoffroy et al., 2013). Previous studies have used this two-layer model to successfully emulate CMIP5 model projections of GSAT and global mean thermosteric SLC to 2300 (Palmer et al., 2018; Yuan and Kopp, 2021). The AR6 emulator ensemble was constrained using four observational targets, including historical GSAT change and ocean heat uptake (Smith et al., 2021). The projected ocean heat uptake was translated to global mean thermosteric SLC using CMIP6-based estimates of expansion efficiency (Fox-Kemper et al., 2021b). The GSAT changes were also used as input for the land-ice contributions to GMSL rise, which were generated with additional emulators applied to suites of coordinated community efforts for the ice sheet (LARMIP-2, Levermann et al., 2020; ISMIP6, Nowicki et al., 2016) and glacier model (GlacierMIP2, Marzeion et al., 2020) simulations carried out for AR6.

Differences in the projected contributions to SLC

In AR5, the assessments of glacier and ice sheet contributions were based on a range of individual models and publications. The only difference in the SROCC projections with respect to AR5 was the reassessment of the Antarctic dynamics contribution, by replacing the AR5 Antarctic scenario-independent ice dynamic projections with scenario-dependent process-based model estimates (Levermann et al., 2014; Golledge et al., 2015; Ritz et al., 2015; Bulthuis et al., 2019; Golledge et al., 2019). This led to a decrease in 21st century GMSL change compared with AR5 for the RCP2.6 scenario, and an increase for the RCP8.5 scenario (medians and likely ranges; Figure 1a,b). However, the scenario-dependence in SROCC may have been amplified because two model estimates did not include accumulation changes (Levermann et al., 2014; Ritz et al., 2015), which are projected to increase with warming and partially counteract dynamic losses (Fox-Kemper et al., 2021a).

For the AR6 projections, statistical emulators were applied to the ISMIP6 and GlacierMIP2 outputs, using the Gaussian process model described in Edwards et al. (2021). For LARMIP-2, results for Antarctic ice sheet dynamics were emulated using an impulse-response function model following (Levermann et al., 2020), augmented by a parametric surface-mass balance model following AR5. There were several motivations for using these emulators: (1) to constrain the projections to the assessed ECS range, an approach that represents a marked change from previous IPCC reports; (2) to be able to make projections across all five illustrative SSP scenarios of AR6, as the ice sheet and glacier contributions were mostly based on CMIP5 RCP scenarios; and (3) to sample modelling uncertainties more thoroughly, estimating probability distributions for the contributions. The use of simple climate models and emulators is a trade-off between a more complete exploration of the uncertainties which can be done due to the computational speed of the emulators (compared to the full ice sheet and glacier models, which are limited by constraints of computing and person time), and the potential biases introduced by the necessary assumptions of a simpler model (Edwards et al., 2021). The Gaussian process emulator performed well for the cumulative change in time, but did not account for temporal correlation, so the rates could not be estimated from the emulator. As a consequence, in contexts where rates were needed, AR6 used simpler parametric emulators, based on approaches used in AR5.

A comparison of the GMSL projections to 2100 in the different reports reveals a number of differences (Figure 1a,b). In the land ice contributions, we see a narrowing of the likely ranges for glaciers (under both scenarios) and the Greenland ice sheet (under SSP5–8.5), and a widening of the Antarctic ice sheet likely range. The latter is wider as it is based on a p-box bounding distribution functions from the ISMIP6 emulator and LARMIP-2 (Table 1), where the presented likely range spans from the lowest 17th to the highest 83rd percentile of the considered methods. The ECS-constrained temperature projections in AR6 (section ‘Updated climate model information and the use of emulators’) used as input to the land ice emulators show a marked reduction in the width of the likely range at 2100 (~0.7 K for SSP1–2.6; ~1.9 K for SSP5–8.5) compared with the 21 CMIP5 models used as the basis of the AR5 sea-level projections (~1.6 K for RCP2.6; ~2.8 K for RCP8.5), which could also be one of the reasons for the reduced width of the glacier and Greenland likely ranges. The glacier range may also be slightly underestimated because each region is emulated independently, which means the projections do not account for covariances in the regional uncertainties apart from those associates with a common dependence on temperature (Marzeion et al., 2020; Fox-Kemper et al., 2021a, Section 9.5). However, the AR5 glacier and Greenland uncertainties were open-ended (≥ 66% ranges) and essentially estimated with expert judgement, at a time of far less information from – and confidence in – these process-based models, so the narrowing range is also consistent with an improving evidence base. The land-water storage contribution is reduced in AR6 compared with AR5 due to the use of a different methodology which now links land-water storage changes to global population under SSP scenarios (Kopp et al., 2014), in combination with a larger negative reservoir impoundment contribution from Hawley et al. (2020).

AR5 used different methodologies for estimating the uncertainties in GMSL (Figure 1a,b) and regional SLC (Church et al., 2013b). In contrast, the AR6 GMSL (Figure 1a,b) and regional projected uncertainties are combined in the same way, with the different contributions all treated as conditionally independent given GSAT, which is an input for the emulator (Fox-Kemper et al., 2021b). The total projected GMSL for SSP1–2.6 has increased in AR6 compared with RCP2.6 projections in AR5 and SROCC, with a similar likely range (Figure 1a), but with different relative contributions of each component. For SSP5–8.5, the AR6 GMSL projections are 4 cm lower than RCP8.5 in SROCC but 6 cm higher than RCP8.5 in AR5 (Figure 1b), due to differences in the model estimates included (from AR5 to SROCC) and in both models used and the methods used to combine the models (from SROCC to AR6) of the projected Antarctic contribution.

The regional projections (Figure 2) show that SLC is spatially highly variable, due to a combination of ocean dynamic changes, gravitational, rotational and deformation (GRD) effects in response to present-day mass changes, and long-term Glacial Isostatic Adjustment (GIA). There is an overall agreement in the patterns between AR5 and AR6. Some differences arise from the vertical land motion (VLM) contribution, which included only GIA in AR5 and also other VLM contributions, such as tectonics, compaction or anthropogenic subsidence, in AR6: compare for instance the larger ratios along the US East Coast ( Figure 2b) to the VLM contribution in Figure 9.26 from Fox-Kemper et al. (2021a). The increased contribution from Antarctica compared to AR5, in combination with the ocean dynamics contribution, leads to a more widespread below-average SLC in the Antarctic Circumpolar Current region.

Sea-level projections outside the likely range

One of the key uncertainties in sea-level projections is the dynamic contribution of the ice sheets (i.e., processes related to the flow of the ice). AR5 assessed the likely dynamical contribution of the Antarctic Ice Sheet by 2100 at −2 to 18.5 cm, but also noted that ‘Based on current understanding, only the collapse of marine-based sectors of the Antarctic ice sheet, if initiated, could cause global mean sea level to rise substantially above the likely range during the 21st century. There is medium confidence that this additional contribution would not exceed several tenths of a metre of sea level rise during the 21st century’. An ice sheet estimate based on SEJ was available at the time of AR5 but this could not be supported by other lines of evidence (Church et al., 2013a). Including the SEJ estimates would have led to an assessment that could not be transparently linked to physical evidence, as the reasoning of the experts involved in the SEJ exercise is undocumented, and it was decided not to use it for the AR5 assessment.

After AR5, following for instance Sutton (2019), low probability estimates were increasingly used in the context of risk assessment and to discuss less likely outcomes for risk-averse users (e.g., Le Cozannet et al., 2017b; Hinkel et al., 2019; Nicholls et al., 2021). SROCC argued that stakeholders with a low risk tolerance might use the SEJ numbers (e.g., their Figure 4.2). Model results including marine ice cliff instability (MICI, Deconto and Pollard, 2016) were not used in the main projections of SROCC because the too high surface melt rates led to an uncertain timing and magnitude in the simulated ice loss. In AR6, a set of low confidence projections was presented (shown in grey in Figure 1a,b) which build on the medium confidence projections. These projections include additional contributions for the ice sheets, estimated using a p-box approach (e.g., Le Cozannet et al., 2017a), considering SEJ (Bamber et al., 2019) together with an improved model-based estimate for Antarctica which included MICI (DeConto et al., 2021). It is important to note that the low confidence ranges represent the breadth of literature estimates available at the time, but that they are not incorporated in the assessed likely ranges.

The AR6 low confidence projections suggest that by 2100, under SSP1–2.6 ( Figure 1a), there is a potential Greenland contribution outside the likely range, based on SEJ. For Antarctica, the medium confidence SSP1–2.6 projections already include a wide range of values, so the impact of SEJ and MICI estimates in the lowconfidence projections is less distinct. Under SSP5–8.5 (Figure 1b), the upper values of the AR6 low confidence projections for both ice sheets are considerably larger than the corresponding medium confidence estimates. This reflects the deep uncertainty in the literature on the Antarctic contribution (see also Box 9.4 in Fox-Kemper et al., 2021a). What is needed to reduce this deep uncertainty is primarily a better understanding of the physical processes. This will lead to more physically-based model projections with larger ensembles, which will allow for a better exploration of the uncertainties.

Comparison of the AR5 model simulations with observations

In the previous section, we discussed ‘how we got here’: the developments that led to the most recent IPCC projections. However, it is also relevant to see ‘where we are’, by comparing the observed sea-level change against sea-level projections for their overlapping period. We evaluate the assessed likely ranges of the AR5 projections (from 2007 onwards, Church et al., 2013a) against the assessed observational time series from AR6 (up to 2018, Fox-Kemper et al., 2021a, Table 9.5), both for the total GMSL and the individual contributions (Figure 3).

For GMSL, Antarctica, Greenland and thermal expansion, the observational timeseries are close to the centre of the projections and the estimated rates of change are highly consistent (Figure 3a,b,c,e). The observed glacier timeseries in AR6 is at the lower end of the projections, even though the observed rates entirely fall within the likely range of the projected rates (Figure 3d). It is worth noting that the AR5 included glaciers peripheral to the Greenland Ice Sheet in the glacier projections (their Table 13.5), which according to the observations in their Table 13.1 adds a contribution in the order of 0.1 mm/yr. In AR6, this was included in the Greenland contribution. To facilitate the comparison, we have included the observed Greenland peripheral glacier estimate in Figure 3d (Glaciers) and subtracted it from the observations in Figure 3c (Greenland), based on linear rates presented in Church et al., 2013a and Fox-Kemper et al., 2021a. In addition, the glacier contributions since AR5 suggest a smaller glacier contribution, both in observations and projections (for the observations: grey dashed line in Figure 3d based on Church et al., 2013a, Table 13.1 shows a higher rate than AR6, for the projections: Marzeion et al., 2015). The observed rate of land-water change is larger than the projected central value, but the observed time series, despite its interannual variability, mostly falls within the projected likely range. The observed rate of change is at the upper bound of the likely range projections (Figure 3f).

Wang et al. (2021a) also evaluated GMSL and regional projections from AR5 and SROCC against different tide gauge and altimetry time series for the period 2007–2018. They found that the GMSL trends for 2007–2018 from AR5 projections are almost identical to observed trends and well within the 90% confidence interval. They also showed significant local differences between observations and models, which could be improved with better VLM estimates and minimisation of the internal variability.

A study by Lyu et al. (2021) focused on ocean warming, with the purpose of constraining projections. They compared the observations of ocean temperature by the Argo array (2005–2019) with model simulations from the CMIP5 and CMIP6 databases. They found that (1) the range of CMIP6 has shifted upwards compared with CMIP5; (2) there is a high correlation between observations and models over the Argo period; (3) the emergent constraint indicates that the larger trend of thermosteric SLC in the CMIP6 archive needs to be taken with caution. This supports the AR6 approach, where an emulator was used to constrain the thermosteric SLC of CMIP6 models with the assessed ECS range (section ‘Updated climate model information and the use of emulators’), leading to thermosteric SLC projections similar to AR5 and the constrained Lyu et al. (2021) projections.

Moving towards local information

AR5 was the first IPCC assessment report to show regional sea-level projections in addition to GMSL projections, by including the effects of changes in ocean density and circulation, GIA and GRD effects (Table 1). SROCC built on AR5 but explored regional changes in sea-level extremes in more depth. In AR6 as a whole, even stronger emphasis was put on regional climate changes and on using regional information for impacts and risk assessment, in particular in Chapter 10 (Doblas-Reyes et al., 2021), Chapter 12 (Ranasinghe et al., 2021) and the Interactive Atlas (Gutiérrez et al., 2021). The IPCC authors and the IPCC Technical Support Unit also collaborated with NASA to develop the NASA/IPCC Sea Level Projection Tool (https://sealevel.nasa.gov/ipcc-ar6-sea-level-projection-tool) to provide easy access to global and regional projections. As the need for more detailed sea-level information is becoming increasingly evident (e.g., Le Cozannet et al., 2017b; Hinkel et al., 2019; Nicholls et al., 2021; Durand et al., 2022), we discuss a couple of potential future research avenues which may help to further improve sea-level projections on a regional to local scale.

High-resolution ocean modelling

Ocean dynamic SLC is a major driver of spatial sea-level variability, which is typically derived from CMIP5 and CMIP6 GCM simulations. However, the extent to which GCMs can provide local information is limited because of their relatively low atmosphere and ocean grid resolutions, which are constrained by computational costs. The typical ocean grid resolution of CMIP5 models is approximately 1° by 1° (~100 km). Although the ocean components of some CMIP6 models operate at a 0.25° resolution, the resolution of most CMIP6 models has not increased much relative to CMIP5, and the CMIP5 and CMIP6 simulations of ocean dynamic SLC show similar features (Lyu et al., 2020). These relatively coarse resolutions may lead to misrepresentations of ocean dynamic SLC, particularly in coastal regions in which small-scale and tidal processes and bathymetric features are important. Increasing the resolution of the GCMs requires significant additional computational resources as well as more explicit modelling of high-resolution processes that are currently parameterized.

As an alternative, GCMs can be dynamically downscaled using high-resolution atmosphere or ocean models. Emerging research demonstrates the value of dynamical downscaling for SLC simulations in coastal regions such as the Northwestern European Shelf (Figure 4; Hermans et al., 2020; Chaigneau et al., 2022; Hermans et al., 2022), the Southern Ocean (Zhang et al., 2017), the Mediterranean Sea (Sannino et al., 2022), the marginal seas in the Northwest Pacific Ocean (Liu et al., 2016; Kim et al., 2021), the marginal seas near China (Jin et al., 2021) and the Brazilian continental shelf (Toste et al., 2018), on both annual and sub-annual timescales. Additionally, dynamical downscaling can offer a framework in which local changes in tides, surges and waves can be resolved in conjunction with time-mean SLC and incorporated into sea-level projections (Kim et al., 2021; Chaigneau et al., 2022), as it allows for modelling changes at higher temporal frequencies. Dynamical downscaling requires GCM output as boundary conditions, which means that the regional solutions due to the explicit modelling of higher resolution processes should always be considered in the context of the GCM model that is driving the regional model. For instance, for the South China Sea, Jin et al. (2021) found that ‘the downscaled results driven by ensemble mean forcings are almost identical to the ensemble average results from individually downscaled cases’. However, more extensive analysis of the uncertainties associated with dynamical downscaling remains to be done. As a result, the dynamical downscaling of ocean simulations has not yet been systematically applied in the context of regional and local sea-level projections.

Figure 4. Ocean dynamic SLC northwest of Europe, as simulated by (a) the CMIP5 GCM HadGEM2-ES and (b) dynamically downscaled using regional ocean model NEMO-AMM7, and by (c) the CMIP5 GCM MPI-ESM-LR and (d) dynamically downscaled, for the scenario RCP8.5 (2074–2099 minus 1980–2005). Figure adapted from Hermans et al. (2020).

Vertical land motion

In addition to the ocean and ice contributions, relative SLC is affected by VLM (Table 1), which may amplify or even dominate the SLC experienced at coastal locations. AR5 and SROCC used GIA models to estimate the VLM contribution to SLC, whereas AR6 based its VLM estimate on the geological background rate at tide gauge stations, derived using the Gaussian Process Model from (Kopp et al., 2014; Table 1). Neither method provides a satisfactory answer, given that the former excludes non-GIA VLM contributions, and the latter requires assumptions regarding the spatio-temporal extrapolation of the tide-gauge derived background rates to areas without tide gauge information by using a GIA model as a prior. AR6, therefore, stated that ‘there is low to medium confidence in the GIA and VLM projections employed in this Report. In many regions, higher-fidelity projections would require more detailed regional analysis’.

Work published after the IPCC AR6 literature deadline has provided new observation-based estimates of VLM for 99 coastal cities based on InSAR observations (Wu et al., 2022) and along the world’s coastlines using GNSS data (Oelsmann et al., 2021). However, even with better observational estimates, significant assumptions are required when extrapolating these into the future. Both AR5 and AR6 assume VLM rates remain constant over time, an assumption that is wrong in regions that are tectonically active (where VLM will be nonlinear and stochastic) or where VLM occurs in response to groundwater and gas extractions (which is strongly dependent on societal choices). A potential solution is to use expanded geological reconstructions of paleo sea level on millennial time scales to constrain long-term average trends (Horton et al., 2018b).

Conclusions and future perspectives

In this overview, we have discussed several aspects of sea-level projections: recent developments in the projections, how they compare against observations, and potential future research directions: ‘how we got here’ (section ‘Key advances in sea-level projections up to IPCC AR6’), ‘where we are’ (section ‘Comparison of the AR5 model simulations with observations’) and ‘where we’re going’ (section ‘Moving towards local information’).

Key differences between AR5, SROCC and AR6 include the use of new climate model information (CMIP6) and the use of emulators to constrain the projections to the AR6 assessment of Equilibrium Climate Sensitivity (section ‘Updated climate model information and the use of emulators’), new information for the different projected contributions to sea-level change (section ‘Differences in the projected contributions to SLC’), and the treatment of projections outside the likely range (section ‘Sea-level projections outside the likely range’).

The likely range projections of GMSL and regional SLC at 2100 show relatively modest changes from AR5 to SROCC and AR6, given approximately equivalent climate change scenarios (sections ‘Updated climate model information and the use of emulators’ and ‘Differences in the projected contributions to SLC’): under RCP2.6/SSP1–2.6 from 0.25–0.58 m (AR5) to 0.33–0.62 m (AR6); under RCP8.5/SSP5–8.5 from 0.49–0.95 m (AR5) to 0.63–1.01 m (AR6). Substantial reductions in the uncertainty of the Greenland and glacier contributions to GMSL at 2100 under SSP5–8.5 for AR6 are counterbalanced by an increase in the Antarctic uncertainty, which leads to relatively small changes in overall uncertainty at 2100 between AR5 and AR6.

In AR6, the explicit inclusion of low confidence projections highlighted the deep uncertainty associated with the dynamical ice sheet contribution (section ‘Sea-level projections outside the likely range’), which was communicated through the use of ‘low-likelihood high-impact’ storylines (IPCC, 2021; Fox-Kemper et al., 2021a). Regional SLC projections based on the low confidence projections were also provided by AR6, but we highlight that more work is needed on understanding and physical modelling of the ice sheet contributions, and on the potential for different regional estimates associated with the partitioning of Greenland and Antarctic ice mass loss.

Our comparison of AR5 projections with observations for the period 2007–2018 shows that the rates of change agree within uncertainties for GMSL and for individual contributions (section ‘Comparison of the AR5 model simulations with observations’), which is in line with previous studies focusing on total sea-level change (Wang et al., 2021a) and the ocean heat uptake contribution (Lyu et al., 2021). Monitoring the projections against observed changes is important as it can help to constrain future projections.

In terms of future developments of sea-level projections (section ‘Moving towards local information’), we highlight the need for dynamical ocean downscaling to represent processes missing in GCMs, such as tidal effects and local currents in shelf sea regions, to better estimate future ocean dynamic SLC. This would also improve simulations of key small-scale processes at the ocean-ice interface that affect the climatic drivers of ice sheets and therefore projections of their future evolution. It would also lead to a better quantification of the effects on mean SLC on, for example, tidal characteristics and wave propagations to understand the potential compounding effects on future coastal flood hazards. A second aspect that is relevant to relative sea-level projections, in particular in low-lying delta regions, is the need for improved VLM observational estimates and projections. This will particularly impact coastal SLC projections, as flood risks depend on (and in some parts of the world are dominated by) the movement of the land in addition to the changes in water level.

In this paper, we have focused on sea-level projections up to 2100. However, it is important to note that sea-level change does not stop in 2100. Currently, projections beyond 2100 are typically based on different methods compared with the projections up to 2100, due to a lack of model simulations and literature. For instance, in AR6 the time series were extended to 2150 assuming constant ice sheet rates post 2100 and the Gaussian process emulators were substituted with parametric fits. Unfortunately, the use of different methods tends to lead to discontinuities in the time series. To fill this gap, we need better understanding and process modelling of the different components, such that consistent methods can be used to generate long-term projections for the next IPCC assessment report and beyond. This will allow investigations of for instance the sea-level response to surface warming overshoot scenarios, or the inclusion of tipping points in sea-level projections (e.g., Lenton et al., 2019). These are only some of the many potential research avenues associated with long-term sea-level projections, all of which are important to investigate given the long-lasting commitment and widespread consequences of future sea-level rise.