The evolution of UK sea-level projections

The methods used to generate process-based global and local mean sea-level projections have evolved substantially over the last fifteen years, including improved process understanding, advances in ice-sheet modelling, the use of emulators and further development of high-end scenarios. During this time, two sets of UK national sea-level projections have been generated as part of the UK Climate Projections in 2009 (UKCP09; Lowe et al 2009) and in 2018 (UKCP18; Palmer et al 2018b). UKCP18 presented local mean sea-level projections for the UK coastline for the 21st century rooted in Coupled Model Intercomparison Project Phase 5 (CMIP5) models and in methods used in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5), with an emulator-based methodology to provide extended projections to 2300 (Palmer et al 2018a; 2020). We compare UKCP18 global and local mean sea-level projections with those presented in the IPCC Sixth Assessment Report (AR6, IPCC ). We find the likely range projections (characterising the central two-thirds of the distribution) are broadly similar at 2150 to within 0.1 m, except at Edinburgh, where the maximum difference is 0.22 m under medium emissions. Differences arise due to higher contributions from sterodynamic processes and the Antarctic ice sheet, and higher or lower vertical land movement, in AR6 compared to UKCP18. We also compare high-end sea-level rise estimates, presented in AR6 and UKCP09, finding reasonable global and UK local agreement over the 21st century. We explore future paths for UK sea-level science considering both user needs for information and developments in modelling capability. Future UK sea-level projections would benefit from updated high-end sea-level rise scenarios which extend beyond 2100 and continued efforts to build understanding of observed sea-level change drivers. Alongside close collaboration with user groups this would enhance the utility of local sea-level projections by UK coastal practitioners and decision-makers.


Introduction
Sea-level rise is expected to cause significant impacts and stresses on the coastal zone, in the UK and worldwide. Global mean sea-level (GMSL) is steadily rising and will continue to rise over the 21st century and beyond, the rate of GMSL rise having increased from 1.3 mm per year (1901 to 1971) to 1.9 mm per year (1971 to 2006) and more recently further increased to 3.7 mm per year (2006 to 2018) (IPCC 2021b). During this latter period, the observed contributions to GMSL change from thermal expansion and land-ice melt were 39% and 44% respectively (Fox-Kemper et al 2021), the rate of ice-sheet loss having increased by four times between 1992 to 1999 and 2010 to 2019 (IPCC 2021b). Despite some progress towards mitigation targets (e.g. United Nations Climate Change Conference of the Parties COP26), some processes of physical sea-level change are 'locked in' due to past emissions (e.g. the long-timescale response of the ocean and ice sheets) (Arias et al 2021), and some degree of worldwide adaptation is inevitable (Haasnoot et al 2021). Coastal communities around the world will face the challenge of adapting fast enough, cost-effectively and within adaptation limits (Haasnoot et al 2021). Firstly, we explore the central likely range of GMSL and LMSL projections. In IPCC reports, calibrated language is used to describe quantified uncertainty using imprecise probability terms, where likely characterises the central two-thirds of the probability distribution, for a given scenario with medium confidence . In AR5, the likely range of sea-level projections was presented as the 5th to 95th percentile range with about two-thirds chance of sea level falling within the presented range, whereas in AR6, the likely range had at least two-thirds probability 'encompassing the outer 17th to 83rd percentiles of the probability distributions considered in a p-box (Le Cozannet et al 2017)' as described in Slangen et al (2023). The central range of UKCP sea-level projections was presented as the 5th to 95th percentile range of the model ensemble, aligning with previous IPCC reports, yet had distinct definitions (see section 2.1.4).

Methodological developments
Secondly, we explore high-end mean sea-level rise, which lies outside of the central likely range but remains physically plausible. High-end sea-level rise was expressed as a High-plus-plus (H++) scenario in UKCP09 and as a high-impact low-likelihood storyline in AR6. The H++ scenario was not updated for UKCP18, and therefore current guidance is still based on H++ from UKCP09. Throughout our study, the term 'high-end' will be used for a generic high-end sea-level rise scenario, whilst 'H++' and 'high-impact low-likelihood' will be used to refer specifically to the scenarios developed and used within UKCP and AR6 respectively.

Evolution of global mean sea-level (GMSL) projections
Since process-based LMSL projections are derived from GMSL projections, any differences in projected GMSL will tend to translate into differences in LMSL. Further differences in the representation of local processes, such as the estimated spatial patterns associated with land-based ice and water changes (e.g. Tamisiea and Mitrovica 2011), could also give rise differences between projections. We therefore focus first on the changes associated with generating GMSL projections, before exploring changes in generating LMSL projections. GMSL change is constructed by combining the contribution of sea-level change from: (1) Global mean thermosteric sea-level rise (GTE); (2) Mass changes from the Greenland and Antarctic ice sheets due to (a) surface mass balance and (b) ice-sheet dynamics; (3) Mass changes from glaciers and ice caps; and (4) Mass changes from land water storage change due to groundwater extraction and reservoir impoundment. Sea-level components associated with mass changes (2 to 4) are referred to as 'barystatic' components .
It is important to note that UKCP09 and UKCP18 GMSL projections have their foundations in the IPCC Fourth Assessment Report (AR4; Meehl et al 2007) and Fifth Assessment Report (AR5; Church et al 2013) respectively, using the same generation of Global Climate Models (GCMs) from the Coupled Model Intercomparison Project (CMIP) framework which underpin each IPCC report (CMIP3 and CMIP5 respectively). Hence, the methodological and model developments at the time will be reflected in UKCP projections. For a detailed comparison of IPCC sea-level projections, the reader is referred to Slangen et al (2023).
2.1.1. Scientific confidence in understanding and projecting sea-level change Over the last fifteen years (between AR4 and AR6), scientific confidence in our understanding of past and future sea-level change has increased through consistent analysis and improved closure of the global energy and sealevel budgets (Frederikse et al 2020, Forster et al 2021, Fox-Kemper et al 2021. Sea-level budget analysis comparing observed total sea-level change with the sum of the observed components of sea-level change has seen important advances in sea-level budget closure within uncertainties on a global (Box 9.4, Fox-Kemper et al 2021) to regional (basin) scale (e.g. Frederikse et al 2016, Frederikse et al 2020) over specific time periods. Local scale sea-level budget closure has typically been challenging due to large uncertainties in observed processes such as ocean dynamics (see section 2.2.1) and VLM which are difficult to model and measure (Fox-Kemper et al 2021), yet improved closure has been made at sub-basin scales  and at tide gauge locations (Wang et al 2021b).
To evaluate the ability of models in simulating global to regional sea-level change, historical model simulations and observations of total and individual components of sea-level change have also been compared. These comparisons have generally had better agreement over the latter half of the 20th century at global (Slangen et al 2017) and regional (Meyssignac et al 2017) scales. In particular, climate models are unable to simulate some observed regional sea-level changes which occur before 1970 .
Agreement between process-based sea-level projections and observations has also improved between IPCC reports. Wang et al (2021a) evaluated AR5 GMSL and regional weighted mean sea-level projections against satellite altimeter and tide gauge observations from 2007 to 2018, finding consistent trends to within the 90% confidence level. Slangen et al (2023) extended this analysis for GMSL, and compared the AR6 observational time series against AR5 projections (2007 to 2017). They found GMSL, GTE, and the Antarctic and Greenland ice sheets showed high consistency on rates of projected and observed sea-level change whilst other individual component trends agreed mostly to within the likely range. Altogether, these findings have installed greater confidence in simulating historical and projected GMSL.

Climate change emissions and socioeconomic scenarios
A key difference between IPCC reports is the evolving climate change emissions and socioeconomic scenarios. These scenarios are important since they underpin the projections produced in the IPCC and UKCP reports and influence fundamental aspects of model simulations, such as climate radiative forcing (Forster et al 2021). AR4 climate models were forced by scenarios from the Special Report on Emissions Scenarios (SRES) (Nakićenović et al 2000). The SRES scenarios were constructed from narrative storylines (including A1, B1) of emissions and socioeconomic developments, taking into account potential energy mixes such as fossil intensive (A1FI) or more balanced energy sources (A1B). Superseding the SRES scenarios, the Representative Concentration Pathways (RCPs) (Meinshausen et al 2011) emissions scenarios forced climate models used in AR5. The RCPs presented a range of greenhouse gas and other radiative forcing levels over the 21st century (2.6, 4.5, 6.0, 8.5). The RCPs did not account for socioeconomic narratives, unlike the five Shared Socioeconomic Pathways (SSPs) (Riahi et al 2017) which forced climate models used in AR6. Baseline SSPs were assessed with RCP radiative forcing levels and matched with the appropriate mitigation targets they could meet, which is reflected in the notation used throughout this study (SSPX-Y, where X is the SSP number and Y is the radiative forcing pathway).
A comparison of emissions scenarios is shown in figure 2(a). Moving between AR4 and AR5, the inclusion of RCP2.6/SSP1-2.6 drives a wider range of possible emissions futures compared to SRES. Whilst SRES A1FI and RCP8.5 show similar carbon dioxide (CO 2 ) concentrations, the SSP5-8.5 scenario has a higher CO 2 concentration towards 2100, despite having the same approximate radiative forcing as RCP8.5. This is related to SSP5-8.5 having a higher CO 2 concentration and lower methane (CH 4 ) concentration mix compared to RCP8.5 (see Chen et al 2021).
The effect of evolving emissions scenarios on median GMSL change using UKCP18 methods is shown in figure 2(b). The GMSL projections under SSP scenarios are from Hermans et al (2021), using the same scenariodependent parameterisation of Levermann et al (2014) for Antarctic ice dynamics as Palmer et al (2018b) (see section 2.1.3), making an appropriate GMSL comparison using UKCP18 methods. The SRES A1B scenario translates into GMSL change which lies between RCP4.5/SSP2-4.5 and RCP8.5/SSP5-8.5, as indicated by the CO 2 concentration time series. On applying the same methods to project GMSL, SSP and RCP scenarios with corresponding radiative forcing only differ by a few centimetres at 2100 (consistent with Hermans et al 2021), making an appropriate comparison for our study. Throughout our analysis, we therefore compare SRES B1 with RCP4.5/SSP2-4.5 (medium emissions) and SRES A1FI with RCP8.5/SSP5-8.5 (high emissions), with RCP2.6/ SSP1-2.6 assigned to low emissions.

Inclusion of ice-sheet dynamical changes
The projected central range of GMSL change at 2100, presented in IPCC and UKCP reports, is shown in figure 3, where projected GMSL in UKCP09 and AR4 are equivalent. Ice-sheet dynamical changes (representing the flow of ice into the ocean) for both Greenland and Antarctica were included in AR5 GMSL projections for the first time, leading to a systematic increase in projected GMSL rise relative to AR4 (Church et al 2013) (∼ 70% increase in the central estimate under high emissions). These were scenario-independent assumed distributions of icesheet dynamics based on observation-based scenarios and literature at the time. UKCP18 included a further update to this additional term: using a parameterisation of scenario-dependent estimates of Antarctic dynamic ice discharge from Levermann et al (2014). Including the Levermann et al (2014) estimates caused the largest impact in the 95th percentile of total GMSL rise (∼ 60% and 90% upper bound increase compared to AR4 under medium and high emissions respectively, dashed black box). A similar approach was adopted by the IPCC Special Report on Oceans and Cryosphere in a Changing Climate

Treatment of equilibrium climate sensitivity
The treatment of equilibrium climate sensitivity (ECS) has evolved between IPCC reports. ECS is an emergent property of climate projections and quantifies the equilibrium surface warming for a doubling of the atmospheric concentration of CO 2 (Forster et al 2021). This surface warming is important for sea-level projections since many components are driven by projections of Global Surface Air Temperature (GSAT). The latest climate models (CMIP6), have on average higher ECS and spread than previous generations (CMIP3 and CMIP5, used in AR4 and AR5 respectively; Forster et al 2020). Hermans et al (2021) showed that sub-setting CMIP6 models based on higher or lower ECS can result in GMSL projections substantially outside the 5th to 95th percentile ensemble range.
In AR5, the 5th to 95th percentile range of the model ensemble had been used to present the likely range of GSAT projections and sea-level projections derived from GSAT. In the absence of additional evidence, this interpretation acknowledged that the CMIP5 model range was not able to capture all uncertainty sources. This same percentile range was used in AR4, however it was not possible to assign a likelihood for sea-level projections . In contrast to previous IPCC reports, AR6 combined CMIP6 model-based ECS estimates with multiple sources of evidence, including observed historical warming, paleoclimate data, physical climate process understanding and emergent constraints. Therefore, AR6 was able to reduce the ECS uncertainty range. As a result, the range of projected warming by climate models used in AR6 was more aligned to that of CMIP5 models (see FAQ 7.3 in Forster et al 2021). For consistency with AR6 climate sensitivity assessment, temperature-dependent component projections of sea-level rise (land-based ice and GTE) were forced by GSAT projections from a two-layer energy budget emulator (Smith et al 2018). AR6 likely range sealevel projections were therefore generally presented as the 17th to 83rd percentile range and included sea-level process uncertainty as well as GSAT uncertainty (Fox-Kemper et al 2021, Caretta et al 2022). The 17th to 83rd percentile range of the Antarctic ice sheet component in AR6 was derived using a p-box approach across multiple lines of evidence, taking a conservative view on the overall uncertainty (Fox-Kemper et al 2021, Slangen et al 2023).
Following previous IPCC reports, the central range of sea-level projections in UKCP was also presented as the 5th to 95th percentile range of the model ensemble. This range was defined in UKCP09 as having a '90% probability', and in UKCP18 as having a 'greater than 10% chance that the real-world response lies outside these ranges', of which the likelihood could not be accurately quantified (Palmer et al 2018b). We therefore present the 5th to 95th percentile range of UKCP projections as the likely range. For a detailed discussion on evolving quantified uncertainty, the reader is referred to Kopp et al (2022).

Land water storage (LWS)
In AR4, the anthropogenic change in Land Water Storage (LWS) could 'not be estimated with much confidence' (Meehl et al 2007), however it was noted that groundwater extraction or reservoir impoundment could have substantial contributions to changes in ocean mass. The LWS component was included in UKCP18 and AR6 projections, where both studies used the same spatial patterns of change for LWS (based on Slangen et al 2014). UKCP18 applied this spatial pattern to the AR5 time series of LWS. In contrast, AR6 introduced calibrated LWS estimates based on statistical relationships of both groundwater depletion and dam impoundment with population change (from the SSP scenarios) and applied a 20% correction to account for retention of water on land (Wada et al 2016) (this latter correction was applied to later studies based on UKCP18 methods, e.g. Harrison et al 2021). Whilst UKCP18 assumed groundwater depletion dominated over dam impoundment in the early 21st century, AR6 applied a correction up to 2040 to account for dam impoundment based on those planned or currently under construction. Therefore, the net driver of LWS change was different between AR6 and UKCP18 in the first few decades of the projections. In AR5, projected contributions to global mean sea-level change were scenario independent, whereas in AR6, LWS projections were explicitly driven by population and exhibited weak scenario dependence, with contributions around 0.01 m higher for SSP3 compared to other scenarios (Fox-Kemper et al 2021).

Evolution of local mean sea-level (LMSL) projections
GMSL rise is not uniform; rates of sea-level change vary on regional to local scales. LMSL is constructed from: (1) 'Sterodynamic' change, referring to changes in dynamic sea level (ocean circulation and density) and including GTE; (2) Barystatic changes (section 2.1), which give rise to changes in the Earth's Gravity, Rotation and solid-Earth Deformation (GRD), which are mapped as geographic 'fingerprints'; (3) VLM, associated on long (centennial to millennial) timescales with GIA and tectonics and on short (annual to centennial) timescales with local subsidence, tectonic events or soil compaction. Globally, VLM may be caused by anthropogenic processes, such as subsidence due to groundwater extraction, or by natural processes, such as volcanic activity, landslides, or earthquakes. Whilst we focus here on mean sea level, changes in mean sea level generally drive changes in sealevel extremes across the world (Fox-Kemper et al 2021). Processes of sea-level variability and extreme sea level including tides, storm surges, waves, and seasonal to multi-decadal modes of variability (e.g. El Niño-Southern Oscillation) would need to be combined with LMSL to model high or low water level events, considering effects such as changing water depth on tides and surge components. These regional to local processes of sea-level change, as well global components, are illustrated in the 'Sea-level Jigsaw Puzzle' (figure 4).

Ocean dynamic sea level
Ocean dynamic sea-level (ocean circulation and density) change is an important contributor to LMSL change, driving dominant spatial variations in sea level at locations away from ice sheets (Meyssignac et al 2017, Wang et al 2021b. From a global perspective, compared to observations, Landerer et al (2014) found a substantial improvement in the ability of CMIP5 models to represent dynamic mean sea level compared to CMIP3 models. CMIP6 GCMs were found to exhibit similar performance, as well as similar regional dynamic features, compared to CMIP5 (Lyu et al 2020). However, regional dynamic sea-level projections exhibit a large spread between models (related to air-sea fluxes and the ocean response) in particular where dynamic sea-level projections are larger (Lyu et al 2020). Despite improvements between CMIP3 and CMIP5 (Church et al 2013), this inter-model spread did not remarkably reduce between CMIP5 and CMIP6 (Lyu et al 2020).
Computational cost scales with the resolution of GCMs, which currently limits the representation of key local-scale processes in the ocean (Hewitt et al 2022). The horizontal resolution of CMIP5 GCMs is typically around 100 km by 100 km, approximately the same as the median resolution of CMIP6 models (Hewitt et al 2022), however model resolution varies considerably across GCMs , Hewitt et al 2022. Around the UK, the relatively coarse resolution leads to limitations in accurately representing shallow shelf seas, noticeably the Irish Sea, connections to the North Sea and the English Channel , Hermans et al 2022. GCMs also do not simulate key processes such as tides and surges. Hermans et al (2020) found unresolved processes in CMIP5 GCMs could lead to differences in ocean dynamic sea level of up to 15.5 cm along the North Sea coastline, compared to higher resolution dynamically downscaled simulations driven by GCMs. Dynamically downscaled information has yet not been included previously alongside IPCC LMSL projections or as part of a coordinated downscaling framework for regional sea-level change, however emerging research has shown dynamical downscaling could have a role in LMSL projections, for example offering insights into seasonal to interannual variability around the Northwestern European Shelf (Hermans et al 2020, Tinker et al2020).
In UKCP09, sterodynamic LMSL change was computed by scaling regional variations in AR4 GTE under SRES A1B to higher and lower emissions scenarios (following Nicholls et al 2011), and taking an average of these geocentric estimates around the UK. A different approach was taken in UKCP18, where sterodynamic LMSL was computed by forming regression slopes between GTE and local sterodynamic sea level from each CMIP5 model, for all UK coastal grid boxes. Due to model resolution limitations, a conservative approach was taken to encapsulate the uncertainty which made no assumptions on the spatial pattern of local sterodynamic change. During the Monte Carlo step, the CMIP5 model and coastal grid box regression were therefore drawn at random and combined with GTE to determine local sterodynamic change for a selected location (Palmer et al 2018b). AR6 fitted a multivariate t-distribution to ocean dynamic sea-level change from the CMIP6 ensemble, from which ocean dynamic sea-level change was drawn and combined with GTE based on a two-layer energy balance model emulator (e.g.

Estimates of contemporary gravitational, rotational and deformational (GRD) patterns
The GRD spatial patterns applied to components of GMSL change are important for LMSL projections, since globally they contribute to marked spatial variations in sea level and can attenuate or even change sign of individual component contributions . In the UK, the GRD pattern associated with the nearfield Greenland ice sheet has a strong gradient, and sea-level change locally is impacted by small changes in the pattern, and by uncertainty whereas the far-field Antarctic ice sheet conversely is less impacted by spatial uncertainty (Palmer et al 2018b). The processes of GRD are well understood (Fox-Kemper et al 2021), for example globally Palmer et al (2020) found low standard deviations between three different GRD estimates, showing that GRD uncertainty is negligible away from locations of ice and water mass changes.
In UKCP09, land-ice estimates were taken directly from AR4 and assumed global uniformity for small landice changes. UKCP18 took a random draw of land-ice GRD patterns from two independent estimates (Spada andStocchi 2007, Slangen et al 2014), using the same geographic mass distributions (Slangen et al 2014), providing an estimate of GRD model uncertainty but not uncertainty in mass change. AR6 projections took a similar approach, and used gravitationally self-consistent GRD solvers to compute LMSL change, driven by spatiotemporal quantities of barystatic change (Slangen et al 2014).

Estimates of vertical land motion (VLM), including glacial isostatic adjustment (GIA)
Changes in the land-ice load imposed on the Earth's lithosphere drive a viscous adjustment of the mantle. From a global perspective, this process of GIA contributes to VLM at and adjacent to the area of land-ice mass change, and in these areas can dominate spatial variation from GMSL change (Meyssignac et al 2017). Due to the redistribution of mass, GIA also affects the geoid, and has a GRD spatial pattern (separate to contemporary GRD estimates, Gregory et al 2019). The contribution to sea-level change from GIA due to geoid changes is an order of magnitude less than the contribution from GIA due to VLM (figure 5). The UK is affected predominantly by VLM induced by GIA due to the melting of the Eurasian and British-Irish ice sheets and collapse of the Fennoscandia forebulge, which leads to uplift in Scotland and subsidence in England, Wales and the Shetlands, impacting relative sea-level change (Palmer et al 2018b).
VLM due to GIA was the only source of local spatial variation in UKCP09, using constant rates taken from Bradley et al (2009) subtracted from a UK average of sterodynamic and barystatic sea-level change. UKCP18 used a regional 15-member observationally-constrained ensemble of GIA (Palmer et al 2018b), and used linear interpolation to extract the local value at each coastal grid point. UKCP projections therefore only considered VLM due to GIA. In contrast, AR6 VLM projections included GIA and other local processes recorded by tide gauge observations (figure 4), projected using a spatio-statistical model adapted from Kopp et al (2014). In the UK, GIA is the dominant VLM process that gives rise to spatial variations in the rate of sea level change (e.g. Hogarth et al 2020), yet some coastal areas in the UK, such as the Fenlands, the Norfolk Broads and the Lancashire lowlands experience ongoing subsidence (de la Vega-Leinert and Nicholls 2008) which is not captured in UKCP18 projections. However, Tay et al (2022) showed that whilst AR6 VLM projections are able to capture the broad scale VLM processes (tectonics and GIA) well, they do not capture full range of local spatial patterns of subsidence and are limited by tide gauge distribution.

Exploratory extended projections to 2300
UKCP18 also presented an exploratory extended dataset of 5th to 95th percentile range projections to 2300 (Palmer et al 2018a(Palmer et al , 2018b, for locations around the UK. These were based on the RCPs and their corresponding extended concentration pathways (Meinshausen et al 2011) and were generated using a physically-based emulator (Palmer et al 2018a). These extended projections were designed for practitionersinterested in longer (centennial) timescale coastal planning. The UKCP18 extended projections were however considered to have much lower confidence compared to the 21st century projections, so whilst they can be used for sensitivity studies, they were not intended to represent the most likely or full range of sea-level rise behaviour post-2100 (Palmer et al 2018a). The two UKCP18 datasets show minimal difference in the likely range total sea-level change to 2100 (Palmer et al 2018b), and so throughout our comparison, we employ the extended dataset for comparative purposes with AR6 projections (available to 2150).

Evolution of high-plus-plus (H++) and high-impact low-likelihood storylines
The UKCP09 H++ scenario and the AR6 high-impact low-likelihood storyline were both used to explore the low probability 'high-end' space for mean sea-level projections. These were mainly designed as an attempt to quantify physical processes related to rapid Antarctic ice sheet discharge, due to limited detailed understanding. Quantifying Antarctic ice sheet discharge is important for the UK, since due to GRD effects, the total sea-level change in the UK is more affected by ice-sheet loss from the far-field Antarctica, rather than near-field Greenland. Around the UK, the contribution of mass loss from the Antarctic ice sheet to LMSL change is roughly 1.09 to 1.16 per unit of GMSL change, whilst the contribution from the Greenland ice sheet is roughly zero to −1.13 per unit of GMSL change.
At the time of UKCP09 development, known ice-sheet model limitations and emerging proxy evidence indicated that future sea-level rise could be much greater than the values presented in AR4. The H++ concept was introduced as part of dynamic adaptation strategies under the Thames Estuary 2100 plan (TE2100) in order to address climate model limitations (i.e. in representing ice-sheet processes). An initial estimate was based on expert judgement and revised under UKCP09 (Ranger et al 2013, van de Wal et al 2022). The H++ scenario under UKCP09 was developed as a static range at 2095 of 0.93 m to 1.9 m for LMSL rise around the UK. In this H ++ scenario, the lower limit of GMSL rise was taken directly from the maximum GMSL rise estimate in AR4, whilst the upper limit was derived from average rates of last interglacial sea-level rise (∼125,000 years ago), where the global mean surface temperature and ice-sheet configuration were similar to present day (Rohling et al 2008). For maximum sea-level rise around the UK, this global upper limit was adjusted for regional deviations due to the effect of ice-sheet load changes on GIA. For the UK lower limit, a scaled ice-sheet discharge contribution was applied to account for recent accelerated ice flow. UKCP09 suggested that H++ could be used for contingency planning (Ranger et al 2013), to build an understanding of potential adaptation limits or a 'worst case scenario'. H++ was not updated for UKCP18, and thus current practitioner advice is still (to date) based on the UKCP09 scenario. The adaptive and salient scientific evaluation process in which H++ was updated within the TE2100 plan has inspired the approach of other high-end estimates (van de Wal et al 2022).
In subsequent years, further research sought to model and quantify rapid and runaway Antarctic ice sheet discharge (e.g. DeConto and Pollard 2016, Edwards et al 2019). AR6 was the first IPCC study to present a highimpact low-likelihood storyline for the 21st century and provided sea-level projections to include low confidence processes associated with ice sheets (Fox-Kemper et al 2021) to indicate sea-level changes that extend beyond the range of medium confidence (expressed in the likely range projections). These were based on Structured Expert Judgement (SEJ, Bamber et al 2019) of future sea-level rise, including contributions from both the Greenland and Antarctic ice sheets, and on numerical modelling of the Antarctic ice sheet carried out by DeConto et al (2021) which explored Marine Ice Cliff Instability (MICI; a self-sustaining ice-loss mechanism triggered by rapid collapse of ice shelves). The low confidence IPCC projections provide risk-averse practitioners with an indicator of the 'deep uncertainty' sea-level changes beyond 2100. Throughout this study, we refer to the 83rd percentile of the AR6 low confidence high-impact low-likelihood projections extending to 2300, which consider MICI, as 'high-end'.

Results
In this section we present the main findings of our study. We compare the likely range time series of GMSL and LMSL projections for four UK locations (Belfast, Cardiff, Edinburgh and London), including a component analysis to investigate the source behind any discrepancies. We also compare high-end estimates as described in section 2.3. Tabulated total projected median and likely range GMSL and UK LMSL change at 2100, 2150, 2200 and 2300 where available, and total high-end sea-level estimates, can be found in Supplementary data (S.2, figures S.2.1, S.2.2, S.2.3 and S.2.4). Figure 6 shows a comparison of GMSL likely range projections presented in UKCP18 and AR6 under low (RCP2.6/SSP1-2.6), medium (RCP4.5/SSP2-4.5) and high (RCP8.5/SSP5-8.5) emissions scenarios to 2150, and where available the corresponding extended datasets to 2300. AR6 low confidence projections are also shown for low and high emissions which effectively capture the widest 17th to 83rd percentile range across multiple lines of evidence (see table 9.11, Fox-Kemper et al 2021). This includes the AR6 assessed 17th to 83rd percentile ranges at 2300 in the absence of MICI (solid bar) and the 83rd percentile high-impact low-likelihood projections including MICI (dashed grey line).

Comparison of global mean sea-level (GMSL) projections
Towards 2150, AR6 GMSL projections show good agreement with UKCP18 projections across all emissions scenarios. UKCP18 encapsulates the AR6 likely range for the period of 2020 to 2150 under low and high emissions scenarios. The central estimates and projection uncertainty are similar, typically within 10% under all emissions scenarios at 2150. A 'step' in AR6 projections at 2100 can cause lower bound estimates under low and medium scenarios to differ by around 20%. Consistently wider uncertainty estimates for UKCP18 may be partly explained by inclusion of a 'model discrepancy' term. This is the difference between the CMIP5 model and two-layer model time series for each variable and is factored into the uncertainty of the emulated projections, as discussed by Palmer et al (2018a).
At 2300, the GMSL projections are also broadly compatible, yet the UKCP18 projection has a substantially narrower range compared to the AR6 (MICI-absent) low confidence assessment, particularly in the upper bound. Under SSP1-2.6, the 83rd percentile of the MICI-inclusive assessment sits beneath the 83rd percentile of SEJ, hence a distinction is not made between the consideration of MICI under low emissions. Under high emissions, from around 2150, the UKCP18 projections are considerably lower than the 83rd percentile of the AR6 low confidence projections which include MICI (see section 3.3).
The GMSL projections to 2300 are useful for illustrating multi-century commitment and the benefits of mitigation action for reducing the long-term rise in sea level. Between 2006 and 2018, the average rate of GMSL rise has been observed at 3.7 mm per year. Extrapolating this rate from 2020 would result in GMSL of around 1 m at 2300, which aligns with the central to lower estimates under all emissions scenario in UCKP18 and AR6 GMSL projections. A GMSL rise of 1 m has also been calculated as the committed sea-level rise by 2300 under pledged emissions until 2030 (Nauels et al 2019). Limiting committed GMSL rise to below 1 m would require drastic and rapid reductions in near-term carbon emissions (Nauels et al 2019). This simple example illustrates how decision-makers and practitioners can benefit from improved scientific confidence based on multiple lines of evidence.
On comparing the individual components which build GMSL projections at 2100 (figure 7), the inclusion of the Antarctic ice sheet dynamics term after UKCP09 drives a significant higher and positive shift in GMSL rise (see section 2.1.3). The uncertainties tend to be systematically smaller in AR6 compared to UKCP18, yet the combined uncertainties are similar. The uncertainty under high emissions is larger, which is expected under  larger forcing and associated climate sensitivity uncertainty. The correlation structure among sea-level components means that the Antarctica ice dynamics component tends to dominate the overall uncertainty, as demonstrated by previous studies (Kopp et al 2014. The uncertainties for this component are very similar for UKCP18 and AR6, which would seem to explain why the total uncertainty is also similar between the two sets of projections. Slangen et al (2023) suggested that the reduction in uncertainty for the glaciers and Greenland components in AR6 compared to AR5 is a reflection of the constrained temperature projections used as input to the land-ice emulators, as well as an improved evidence base. Figure 8 shows an evolving comparison of LMSL likely range projections from UKCP09, UKCP18 and AR6, generated at representative tide gauge locations in the UK. Projections for nearby tide gauge locations have been used to compare capital city locations which make up the UK: London (Sheerness), Edinburgh (Leith), Belfast (Bangor for UKCP09 and UKCP18; Belfast for AR6) and Cardiff (Newport for UKCP09 and UKCP18; Hinkley Point for AR6). AR6 nearest tide gauges differ slightly due to current availability on the NASA-IPCC sea-level projection tool, however Hinkley Point to Newport and Belfast to Bangor are around 40 km and 20 km apart respectively, and hence we assume the differences in location have minimal effects in our comparison.

Comparison of UK local mean sea-level (LMSL) projections
The comparison shows UKCP09 projections tend to be consistently lower for all locations, consistent with GMSL projections. UKCP18 central estimates are generally 30% (30% to 45%) higher than UKCP09 under high (medium) emissions for UK locations at 2100. There is generally a high level of agreement between AR6 and UKCP18 for locations around the UK. Central estimates and uncertainty ranges across all emissions scenarios at 2100 and 2150 typically agree to within +0.1 m. The largest differences between AR6 and UKCP18 are exhibited for Edinburgh, with AR6 central estimates that are systematically larger to a maximum of +0.15 m at 2100 and +0.22 m at 2150 under medium emissions (RCP4.5/ SSP2-4.5).
As mentioned in section 3.1, a noticeable feature exists in AR6 projections, where a 'step' in the time series can be found around 2100. This is more pronounced for the higher emissions scenarios. In comparison, UKCP09 and UKCP18 projections run smoothly across 2100. This feature in AR6 appears to be related to the change in methods for land-ice components at 2100, as documented in table 9.7 of Fox-Kemper et al (2021). Up to 2100, Gaussian Process emulators were used to ensure inter-scenario consistency for the glacier and ice-sheet projections, where inputs include the emulated GSAT projections (Edwards et al 2021). Beyond 2100, land-ice components were projected using different methods, such as extrapolating parametric model fits to ice-sheet model projections (Fox-Kemper et al 2021). This direct change in methods is required, since ice-sheet projections in CMIP6 only extended to 2100 (Nowicki et al 2016). This 'step' feature could be problematic for sea-level projection users (see section 4). In comparison, the UKCP18 extended (emulated) dataset runs smoothly across 2100.
A component comparison for UKCP18 and AR6 LMSL projections at 2150 can be found in figure 9. In general, differences between individual components arise yet generally cancel out for total LMSL change, where AR6 glacier and LWS components are generally lower compared to UKCP18, and AR6 sterodynamic, AIS and GIS components are generally higher compared to UKCP18. The VLM is generally lower in AR6 compared to UKCP18, except at Edinburgh, where it is higher. The largest spatial variation appears to arise from the sterodynamic and VLM components. Under medium emissions, the difference at 2150 in the sterodynamic component is atypically large (+0.15 m) at Edinburgh compared to other locations (around +0.1 m), whilst the Antarctic ice sheet component difference is consistent with other locations (typically within 0.1 m), Edinburgh also exhibits the largest VLM difference (+0.07 m) compared to other locations. Generally, differences in the sterodynamic component between studies could relate to the limitations of CMIP models in UKCP18 to resolve shallow shelf seas processes, whilst differences in the VLM component could be attributed to the different processes captured, as well as impact of different GIA model resolutions on the steep GIA gradient in this region. In UKCP09, spatial variability between individual locations is due to the VLM component only. A comparison of VLM between all studies can be found in Supplementary data (S.1, figure S.1.1), showing VLM uncertainties are generally underestimated.
In absolute terms, at most locations, the difference between AR6 and UKCP18 is minimal compared to the total sea-level rise at 2100 and 2150, considering the requirements of practitioners to prepare for sea-level rise under medium to higher emissions scenarios in most use cases (see section 4). Additional lines of evidence, including higher resolution modelling, as well as maintenance of high-quality observations, such as Global Navigation Satellite System (GNSS) co-located tide gauges, would help refine our understanding of sea-level change drivers in this region. Figure 10 compares the UKCP09 H++ scenario range at 2095 (grey dotted range) and the AR6 high-impact lowlikelihood storyline (83rd percentile, red dotted line) for the global and London mean sea-level scenarios. The lower bound of the global H++ scenario is equivalent to the 95th percentile of GMSL presented in AR4, however appears to align with the 83rd percentile of AR6 SSP1-2.6 GMSL rise. This is related to the inclusion of ice-sheet dynamical processes, as discussed in section 2.1.3.

H++ and high-impact low-likelihood storylines
For comparison, AR6 likely range projections are shown to 2150 alongside the UKCP18 sea-level change at 2150 (solid bars) for corresponding low and high emissions scenarios, for GMSL and London local mean sealevel change. We can see that these sea-level scenarios show substantially higher sea-level rise than the likely range projections, and also show agreement up to 2100. However, moving beyond 2100, the AR6 storyline rapidly exceeds the UKCP09 H++ range, more than doubling in sea-level change in around 50 years under high emissions (+125%). This has implications for UK decision-makers and practitioners planning under high-end sea-level rise scenarios.

Discussion
As an island nation with an urgent need to tackle increasing coastal climate risks, the requirement in the UK for LMSL information by practitioners and decision-makers (collectively referred to here as 'users') is widespread, diverse, and contextual. Robust adaptation analysis will require closing the gap between sea-level science with the information needs of decision-makers (Betts and Brown 2021, Palmer et al 2021, see Supplementary data S.4), which can be addressed through collaboration and co-development in coastal climate services UKCP18 held a user engagement process in its development which included participants with different levels of experience and expertise in using climate change information (see Supplementary dataS.3). This engagement involved consulting with key user groups and demonstrating how sea-level projections could be applied (Walkden and Longfield 2018).
In the UK, environmental regulators such as the EA (England); Scottish Environment Protection Agency; Natural Resources Wales and Department of Agriculture, Environment and Rural Affairs are the direct 'superusers' of UKCP LMSL projections, which inform their regulatory requirements on coastal erosion and flood risks, including sea-level allowances (e.g. EA 2022b). Downstream user groups follow their guidance, including those involved in coastal and flood plain infrastructure development, the reinforcement or installation of major barriers and defensive infrastructure (e.g. Thames Barrier, Ranger et al 2013), or long-term coastal planning for vast stretches of the UK's coastline (e.g. Shoreline Management Plans, CCC, 2018). Other direct user groups are involved in adaptive capacity efforts, building scientific or community understanding, for example incorporating projections into coastal risk research (e.g. Perks et al 2023) or using projections to develop status report cards, summaries and reviews (e.g. MCCIP, Horsburgh et al 2020). In addition, user groups include private landowners and non-governmental organisations, who use LMSL projections directly to establish their own guidance, for example in habitat creation and coastal restoration or heritage sites (e.g. RSPB, Miles andRichardson 2018, or ABPmer 2023). In addition, port operators already have a high number of planned, implemented, and potential coastal adaptation strategies (Jenkins et al 2022) to continuously respond to and prepare for the risks to infrastructure and public health from high winds and major storm events, requiring multi-decadal sea-level information (Simm et al 2021). Each user group outlined above will have different decision-making contexts, risk tolerances, and hence could benefit from different types of information as determined through consultation with climate information providers . Here, we discuss the common priorities across UK user groups which have been established through user engagement, and whether these can be met by current and future scientific capability.
UK users have shown interest in continuous, multi-century LMSL projections. For example, UKCP poll results showed over 50% of users had medium to highest interest in projections beyond 2100 and over twothirds showed high interest in location time series projections (see Supplementary data S.3, figure S.3.2; figure S.3.3). This information is important for considering infrastructure and asset lifetimes (including nuclear facilities), studying coastal erosion, and planning for flooding Supplmentary data S.3). The EA, recognising this requirement, sponsored UKCP18 projections to extend to 2300 (Palmer et al 2018a(Palmer et al , 2018b. Current EA guidance requires flood risk assessments by local authorities and developers to plan for LMSL rise up to 100 years in the future (e.g. 2025 to 2125; EA 2022b). On even longer timescales, nuclear facility operators Figure 10. IPCC AR6 central (likely) range mean sea-level projections under the SSP1-2.6 (low emissions) and SSP5-8.5 (high emissions) scenarios for global mean sea level (left) and for London (right). The high-impact low-likelihood storyline under high emissions is shown by the dotted line (83rd percentile). The UKCP18 RCP2.6 (low emissions) and RCP8.5 (high emissions) 5th to 95th percentile ranges (equivalent to AR6 likely range) at 2150 are shown for comparison to the right of the AR6 time series. The UKCP09 H++ range for high-end sea-level rise at 2095 is also shown for the global scenario and for London. All projections are expressed relative to the 1995-2014 baseline, where UKCP18 and UKCP09 projections have been adjusted for consistency with the IPCC AR6 baseline, using a global correction of −0.03 m (see Supplementary data S.5 for information on baseline adjustments). need to consider LMSL rise for the whole facility lifetime (around 160 years) including operation and decommissioning, due to their location along coastlines or tidal estuaries. UKCP user groups showed highest interest in medium to high emissions scenarios (RCP4.5 and RCP8.5, Supplementary data figure S.3.1) and it is such significant infrastructure and assets which require vulnerability testing under the highest emissions scenarios and high-end sea-level rise, to plan and prepare for the most severe yet plausible scenarios (EA 2022b, Office for Nuclear Regulation, ONR, and EA 2020). Results from our comparison may provide users with confidence in using the UKCP18 extended (2300) projection dataset for assessing multi-century LSML rise due to consistency between UKCP18 and AR6 in the overall sea-level hazard profile to 2150. However, new sea-level information is provided in the AR6 multi-century, high-end sea-level projections. To meet user needs, future UK sea-level projections would benefit from updating H++ to provide high-end scenarios which extend continuously beyond 2100. Top-down guidance on high-end sea-level rise for the UK will need to provide clarity on the recommended dataset, the lines of evidence assessed, and how the science of high-end estimates has evolved (Nicholls et al 2021, Simm et al 2021. Adaptive planning approaches and flexible infrastructure designs are recommended in super-user guidance to handle uncertainty (e.g. EA 2022a, ONR and EA 2020), allowing long-term plans to change as new scientific evidence emerges (e.g. Haasnoot et al 2013). This requires salient communication of evolving science, including projection dataset updates. The TE2100 plan is an example of adaptive planning with salient evaluation and revision of the H++ scenario (Ranger et al 2013, EA 2022c, an approach which has inspired successive high-end studies (van de Wal et al 2022). However, practitioners using sea-level projections in the TE2100 plan have found it hard to navigate 'jumps' in time series, such as when the projection baseline is updated Supplementary data S.3). The 'step' feature arising in AR6 projections around 2100, a methodological artefact, may therefore cause similar issues for users, who may naturally interpolate to smooth the projection time series. Moreover, it may appear the science does not have consensus on how sea-level rise will develop across this time horizon. If guidance is updated to include AR6 projections, it should also address this 'step' feature and its interpretation, ensuring clear and consistent communication. Future IPCC projections would benefit from providing continuous, smooth projections focussed to 2200 and to 2300 to cover key planning horizons, which would need to be agreed on developing the climate modelling framework.
The current format of LMSL projections (provided in metres or millimetres per year) restricts their utility for some UK user groups in terms of making clear the socioeconomic threat posed by future sea-level rise . Despite being the dominant driver of changes in sea-level extremes (Howard et al 2019), mean sealevel rise appears to carry low salience compared to the impacts of extreme flooding and erosion events . Improving the awareness of this connection could lead to better decisions towards strengthening coastal resilience. In addition, users assessing coastal impacts may require information on higher spatiotemporal resolution to understand sea-level variability (e.g. to understand the effects on intertidal marine ecosystems, Hermans et al 2022) and to consider interactions between different drivers (e.g. for compounding flooding, inundation or erosion events). Many users need sea-level information in its final form due to capacity (e.g. the provision of total water levels at the coastline for a given storm surge and waves for a given return period). However, translation of LMSL rise to coastal impacts requires additional modelling to assess changes in, for example, coastal flood hazard, coastal geomorphology, and other socioeconomic drivers (CCC 2018). Facilitating timely impacts analysis based on LMSL projections means we need to find effective ways to streamline the results to impacts analysts, in both the academic and commercial sectors.
Systematic modelling limitations may restrict whether resolution requirements can be met by current scientific capability. GCMs are currently unable to resolve many sea-level relevant processes which act on small spatial scales, such as shelf regions and ice-shelf cavities important for studying sudden Antarctic ice sheet mass loss (Hewitt et al 2022), an important contributor to UK LMSL uncertainty. In CMIP5 and CMIP6 GCMs, vertical resolution limitations result in systematic errors in representing shallow shelf seas (e.g. the Northwestern European Shelf), whilst coarse horizontal resolutions mean that some GCMs have a closed English Channel, affecting circulation patterns and hence dynamic sea-level change around the UK (Hermans et al 2020). Refining model spatial resolution could help users concerned with any differences between UKCP18 and AR6 projections (e.g. Antarctic ice sheet and sterodynamic components in Edinburgh LMSL projections). A global representation of kilometre-scale ocean processes may take some time (Hewitt et al 2022), however advances in regional to local higher resolution modelling may help meet user needs. For example, kilometre-scale modelling coupling the atmosphere, ocean and land modelling systems may open insight into UK compound extreme events (e.g. UK Environment Prediction activity, Lewis et al 2019). In addition, dynamical downscaling of GCMs offers insight into seasonal to interannual sea-level variability at higher spatial resolution and could be combined with other sea-level components to form more comprehensive, higher spatiotemporal resolution LMSL projections or feed into impact studies , Hermans et al 2022.
Developing LMSL projections using a collaborative approach between scientists and super-users could help close the information gap for a large number of users. When UKCP18 user engagement was carried out, the model experimental setup for UKCP18 projections had already been established and so consultation focused on informing scientific outputs and preparing users for new information (Supplementary data S.3). Improvements could be made through better understanding how sea-level projections are used, rather than asking specifically what users want, for example when discerning their risk tolerance . A collaborative approach would ensure super-users are involved from the beginning of projection development, working together towards experimental design, and interpreting results through to communication and data access. Since UK environmental regulator super-users are also involved in impact studies (e.g. flood inundation modelling), this would potentially yield a more systematic assessment of sea-level related risks. Having wider knowledge on the access, interpretation, and use of sea-level projections could ensure that downstream users are not two levels removed from sea-level expertise. Whilst illustrated in the context of the UK, this collaborative approach could be a model for other countries looking to establish effective LMSL projections for coastal adaptation. However, such an approach could face challenges due to the super-user organisation's own capacity, strategy, and staff turnover. Moreover, utilising this approach alone would not represent the full diversity of users and their needs, for example where government agencies do not have responsibility for existing privately owned assets. Whilst the UK public would indeed first seek climate change information including sea-level rise from environmental regulators, scientists and the UK government (Steentjes et al 2020), the use of climate information in a policy and regulatory environment can be very different. Effective engagement across user groups will therefore need to be maintained to maximise the utility of sea-level information.
Close collaboration with user groups and super-users can also see co-development of event-based sea-level storylines (Shepherd et al 2018, Betts and. These pathways of plausible future events are being increasingly used to represent the full climate change and sea-level risk landscape, including multi-century highend sea-level estimates (Box 9.4, Betts andBrown 2021, Fox-Kemper et al 2021) and can be used to explore components which are not well captured in many regions of the world (e.g. VLM). As an illustrative example, one could establish different storylines with contributions of components to high-end sea-level rise to meet different risk tolerance levels. These could be further linked to explore cascading risks (Lenton et al 2019), such as accelerated ice-sheet mass loss leading to Atlantic Meridional Overturning Circulation (AMOC) shutdown (e.g. Bakker et al 2016). The AMOC is important since it is related to coastal sea-level variability (e.g. Little et al 2019) and a shutdown could affect extreme sea level through changes in storminess in Europe (Jackson et al 2015). Through collaboration, storylines could be created which best fit information requirements or summarised as a matrix of possibilities for downstream users to explore, depending on their risk tolerance and information required (van Vuuren et al 2014, Stammer et al 2019, Dayan et al 2021. As part of adaptive planning or dynamic decision-making, it is essential to monitor the real-world sea-level trajectory to monitor GMSL, LMSL and component sea-level change (e.g. Met Office sea-level dashboard) in the context of a storyline. It would also be highly beneficial to develop 'early warning' indicators for key ice-sheet dynamic processes, which remain the dominant uncertainty on centennial timescales.

Conclusions
Our results have shown that UKCP18 GMSL and LMSL likely range projections are consistently higher over the 21st century compared to UKCP09 projections. This can be explained by two main scientific developments between UKCP09 and UKCP18. Firstly, the uptake of RCP emissions scenarios led to a wider range of possible sea-level futures compared to those driven by SRES emissions scenarios. Secondly, the inclusion of ice-sheet dynamical processes resulted in substantially higher sea-level change under equivalent emissions scenarios. In contrast, UKCP18 and AR6 likely range projections are broadly consistent at 2100 and 2150 to within 0.1 m, except at Edinburgh where AR6 projections under medium emissions are up to +0.22 m higher, due to differences in the AIS, sterodynamic and VLM components. Since practitioners generally plan for sea-level change under medium and higher emissions scenarios, this difference is minimal in absolute terms when considering total projected sea level towards 2100 and 2150. Despite modelling and methodological advances between UKCP18 and AR6, pragmatically our results indicate the science underpinning UK likely range projections has largely stabilised. From a practitioner's perspective, any decisions based on UKCP18 LMSL projections are expected to be consistent with AR6 projections, since the overall sea-level hazard profile has not changed substantively.
We have also demonstrated that UKCP and AR6 high-end GMSL and UK LMSL estimates are compatible for the 21st century. However, projected sea-level rise in the AR6 high-impact low-likelihood storyline between 2100 and 2150 almost doubles the H++ scenario range presented for 2095. High-end sea-level information beyond the 21st century has not been provided in UKCP reports, yet has been requested, and therefore guidance which considers this higher level of risk should be updated, ensuring clear communication to users. Whilst our comparison finds UKCP18 and AR6 projections to be broadly similar globally and at UK local scales, this may not hold true for projections based on UKCP18 methods in other worldwide regions , Allison et al 2022 where the sea-level rise risk landscape and user priorities may be vastly different. Understanding the drivers of observed LMSL change using high-quality observations, including the maintenance of tide gauges with co-located VLM measurements (e.g. GNSS), will remain a priori ty over the coming decades . This will refine our knowledge of the processes leading to substantial sea-level variability around the UK coastline. Alongside advancing modelling capability and close collaboration with user and super-user groups, and deep, proactive engagement with the local community (CCC 2018, Lawrence et al 2019), this will ensure our scientific understanding is translated into robust sea-level rise information and user-relevant impacts metrics to inform risk-based coastal decision-making.