Mean Sea Level (MSL) represents the long-term average height of the sea surface at a given location, calculated over a period sufficient to smooth out short-term variations such as tides, storm surges, and weather (Hünicke et al., 2015; Pugh and Woodworth, 2014). Globally, MSL has risen at approximately 3–4 mm yr⁻¹ over recent decades, with satellite altimetry showing ∼ 3.3 mm yr⁻¹ over 1993–2021, while shorter-term estimates of ∼ 4 mm yr⁻¹ over 2006–2016 reflect, in part, interannual variability (Hamlington et al., 2024; Llovel et al., 2023). Evidence further indicates an accelerating trend, from ∼ 2.1 mm yr⁻¹ in the early 1990s satellite record to ∼ 4.5 mm yr⁻¹ in recent years (Hamlington et al., 2024), although short-term variability contributes to interannual fluctuations. However, tide-gauge sea-level observations are influenced not only by changes in ocean volume but also by regional processes and vertical land motion (VLM). They therefore record relative sea level (RSL), representing changes in sea level relative to the local land surface rather than absolute ocean volume changes (Oelsmann et al., 2024). Understanding long-term sea-level changes from tide-gauge and archival data is essential for quantifying RSL trends and interactions with tides, storm surges, and extreme events, providing critical information for coastal planning and adaptation at global and local scales (Talke et al., 2018). However, the magnitude and expression of these changes are strongly influenced by regional and local factors. In Ireland, coastal cities such as Cork and Galway are highly vulnerable to flooding. In 2009 a major flood event in Cork led to around EUR 100 million of damages to commercial and residential property (Olbert et al., 2017). Annual economic damages from coastal flooding in Britain are currently estimated at approximately GBP 360 million and are projected to increase two- to threefold by the end of the century (Haigh et al., 2022). Approximately 40 % of Ireland's population lives within 5 km of the coast, including major cities (Wall et al., 2020), making coastal areas highly vulnerable. Given Ireland's similar coastal exposure and geographical position, it is also likely to face significant costs and challenges.

Ireland has a modern network of over 40 tide gauges through the National Tide Gauge Network (NTGN), installed in the 2000s and managed by the Office of Public Works (OPW) and the Marine Institute (MI) (Cámaro García et al., 2021). However, only a few sites, including Malin Head, Dublin Port, and Belfast Harbour, provide long-term (> 40 years) records, which are incorporated into the Permanent Service for Mean Sea Level (PSMSL) dataset (Holgate et al., 2013). Digitized historical series are concentrated in the north and east of the island, including Malin Head (1958–2002; Holgate et al., 2013, extended to the present with MI and OPW data), Dublin Port (1938–present), and Belfast Harbour (1901–2010; Murdy et al., 2015, Fig. 1), whereas records along the south and west coasts are sparse and often of relatively short duration. This uneven geographic coverage makes it difficult to detect regional trends in tide-gauge-derived MSL and RSL, as large portions of the coast remain unmonitored. A minimum of 40 years is generally required to robustly determine trends in tide-gauge-derived sea-level records (Hogarth et al., 2021). Recent work in Britain shows the benefits of systematically recovering and integrating historical tide-gauge data into a unified framework, yielding robust long-term RSL reconstructions, reduced uncertainties, and improved detection of long-term accelerations (Hogarth et al., 2021). Similar historical data rescue efforts across multiple Irish sites are therefore required to improve spatial and temporal coverage of sea-level observations.

Tide-gauge-derived RSL changes are influenced by variations in both ocean level and VLM. Land levels can change due to geological processes, of which Glacial Isostatic Adjustment (GIA) is the primary globally modelled component (Tamisiea and Mitrovica, 2011; Bradley et al., 2011). Across Ireland, RSL change varies markedly, with lower rates in the northeast and higher rates in the southwest, primarily due to GIA (Bradley et al., 2011). During the Last Glacial Maximum, the thickest ice extended from Scotland into Northern Ireland, thinning toward the southwest (Scourse et al., 2021). Following deglaciation, collapse of the forebulge caused land subsidence in the southwest and uplift in the northeast, producing spatial contrasts in RSL trends, including localized sea-level fall in parts of the northeast. Quantifying these impacts is complicated by the range of outputs among available GIA models, including variability within individual subfamilies. Understanding past ice-sheet dynamics is therefore critical not only for explaining historical RSL variations across Ireland but also for refining GIA models, thereby enhancing predictions of future sea-level rise (Clark et al., 2022; Bradley et al., 2023). This is particularly important on the island of Ireland, where regional differences in land movement and sea-level response make accurate projections essential for coastal planning. GIA models differ due to variations in ice-sheet histories, Earth rheology assumptions, and differences in spatial and temporal resolution and calibration approaches (Bradley et al., 2023). GIA models, such as GLAC1D, use physics-based, glaciologically self-consistent ice-sheet simulations (Tarasov et al., 2012), whereas the ICE-6G GIA model is tuned to fit global RSL and crustal motion observations (Peltier et al., 2015). Regional Bradley-type GIA models, including those based on the BRITICE-CHRONO ice-sheet reconstruction of the British–Irish Ice Sheet, are specifically calibrated to British–Irish RSL and GNSS (Global Navigation Satellite System) datasets (Bradley et al., 2023). These differences strongly influence predicted VLM rates, as demonstrated by studies showing high sensitivity to upper-mantle viscosity and uncertainties in deglaciation histories (Bradley et al., 2011, 2023). Consequently, modelled uplift rates can vary substantially, meaning the choice of GIA model can affect corrected sea-level trends.

In this paper, we aim to integrate sea-level records from across Ireland, including the south and west where observations are sparse, to improve understanding of historical RSL changes. In this study, long-term RSL variations are reconstructed from tide-gauge observations using a hierarchical framework that separates regional signals, site-specific linear components (including GIA-related effects), and non-linear local variability.

The objectives are to: i.

Integrate and analyse new records from the south and southwest to investigate regional RSL variability in areas with limited long-term observations.

All sea-level datasets underwent rigorous validation prior to inclusion in the sea-level reconstruction. To ensure methodological consistency, observations from modern datasets were assessed for data completeness and gaps, with the quality-control approach informed by procedures used within the UK tide-gauge network. UKCMF methodology identifies gaps greater than 4.1 hours in the primary observation channel and recommends excluding months with substantial data gaps from monthly mean sea-level calculations (UK Coastal Monitoring and Forecasting, 2014). Informed by these quality-control principles, the study-specific methodology used in this analysis was as follows: observations were assessed on an hourly basis to identify missing hours. Days with more than 4 missing hours were excluded from monthly calculations, and months with 15 or more such days were omitted. For months passing these quality-control criteria, daily mean sea levels were calculated from the available observations on valid days, and the monthly mean was calculated as the mean of the valid daily means.

Tidal constituents were analysed for all high-resolution datasets where temporal resolution allowed, with particular focus on the dominant semi-diurnal M2 component. M2 amplitude and phase lag were used as primary indicators of record quality, as they are highly sensitive to timing errors, datum inconsistencies, and instrument-related issues, and therefore provide a robust measure of consistency in long tide-gauge series (Pugh and Woodworth, 2014). The M2 tidal constituent amplitude and phase were recomputed using the oce package (Kelley et al., 2022). Where M2 amplitudes or phase lags deviated from the rest of the dataset, monthly M2 values were inspected to identify and flag poor-quality observations. In addition, interannual variability in M2 parameters was assessed, with temporal evolution of both amplitude and phase used to assess datum stability and instrumental consistency across the record. At Belfast, as an example, values falling outside the observed M2 amplitude range (∼ 1.13–1.24 m) were flagged as potential data-quality issues. Similar station-specific assessments were undertaken for the other records. Clearly erroneous spikes (i.e., physically unrealistic or corrupted values) were first identified and removed from the monthly MSL records. A subsequent quality-control step applied a threshold of ±4 standard deviations (SD) from the station mean to identify potential outliers within the monthly MSL values. Values exceeding this threshold were assessed before exclusion across all sites. Such exceedances were rare and were interpreted as likely instrumentation or data-quality issues rather than physically plausible meteorological variability (e.g., storm surges), allowing removal of gross observational errors while preserving realistic variability within the records.

In addition to the procedures described above, selected records required station-specific corrections, including adjustment of timing offsets, removal of poor-quality intervals, and supplementation with nearby tide-gauge records where appropriate. These procedures were undertaken to maximise record continuity while maintaining consistency in datum and tidal characteristics. Full details of station-specific quality-control decisions, omitted periods, and supplementary datasets are provided in the Supplement. Corrected timeseries plots and quality control assessments for the principal tide-gauge records are presented in Figures S4–S11, while Table S2 summarises omitted periods, incorporation of supplementary records, and exclusion of potentially erroneous data. Harmonic tidal analysis (including M2 amplitude and phase) is also used as a quality-control metric for assessing record consistency. Modern tide-gauge datasets obtained from the NTGN (MI and OPW) underwent additional quality control, including assessment of M2 tidal characteristics, statistical outlier detection, and evaluation of datum consistency between overlapping records. Further information on these procedures, including the Arklow–Howth datum adjustment and validation of modern datasets, is provided in the Supplement. The six composite tide-gauge records, supplemented by nearby modern stations, provide the longest continuous observational coverage in northern and eastern Ireland (Malin Head, Belfast, and Dublin Port), whereas the south-western records (Cork, Tarbert, and Galway) consist of more sporadic timeseries that include data gaps (Fig. 2). Independent validation of the tide-gauge records against satellite altimetry was also performed to assess consistency over the satellite era (1993–present); these comparisons are shown in the Supplement (Fig. S12 and Table S3).

The reconstructed relative sea level (RSL) record (Fig. 8) reflects the combined influence of the regional signal, site-specific linear GIA-related contributions, and non-linear local variability across the Irish tide-gauge network. The regional component represents the dominant contribution to the network-mean RSL signal, while the non-linear local component remains comparatively small, indicating that site-specific departures from the broader spatio-temporal pattern are limited. The linear GIA-related component introduces spatial variability between sites consistent with the applied GIA priors, with inferred deviations reflecting differences between the prescribed GIA rates and the modelled site-specific trends.

Sites with shorter observational records, particularly Tarbert, Cork, and Galway, show increased uncertainty when the reconstruction is extended back to 1925, as reflected by wider Bayesian credible intervals (CIs). All sites show increasing RSL rates throughout the reconstruction period, with the increase becoming increasingly evident after the 1970s and particularly pronounced during the 2000s.

Mean modelled RSL rates across the Irish tide-gauge network show an overall increase from the late twentieth century into the 2000s, indicating acceleration in RSL change in recent decades. Here, mean modelled RSL represents the combined influence of the regional, linear GIA-related, and local model components. Instantaneous rates represent the modelled temporal rate of change in the smoothed NI-GAM RSL reconstruction and are reported for individual modelled years. In contrast, period-mean rates represent the average rate of change over a specified interval. The 95 % CI widen towards the end of the modelled period (Fig. 9), reflecting reduced observational constraints near the temporal boundaries of the reconstruction. For 2024, instantaneous RSL rates are estimated at 3.89 mm yr⁻¹ (95 % CI: 2.26–5.56) for Belfast, 4.38 mm yr⁻¹ (95 % CI: 2.71–6.05) for Malin Head, 4.24 mm yr⁻¹ (95 % CI: 2.65–5.86) for Dublin Port, 4.28 mm yr⁻¹ (95 % CI: 2.66–5.88) for Cork, 4.41 mm yr⁻¹ (95 % CI: 2.75–6.02) for Tarbert, and 4.37 mm yr⁻¹ (95 % CI: 2.73–5.99) for Galway.

Long-term mean modelled RSL rates vary geographically across the Irish tide-gauge network. These rates provide a basis for comparing spatial differences in long-term RSL change between sites, with the mean modelled rates for each tide-gauge site summarised in Table 3. Rates vary across the northern sites, with Malin Head exhibiting a mean rate of 1.28 mm yr⁻¹ (95 % CI: 0.86–1.73) and Belfast showing a lower mean rate of 0.73 mm yr⁻¹ (95 % CI: 0.42–1.05) over 1925–2024. The lower rates are consistent with spatial variations in GIA-related vertical land motion across Ireland, with greater relative uplift in the north contributing to lower observed RSL rise. Dublin Port, located on the east coast, shows a mean rate of 1.12 mm yr⁻¹ (95 % CI: 0.83–1.41) over 1925–2024.

Higher mean rates are observed at the southern and western sites, with Tarbert showing the highest rate at 1.43 mm yr⁻¹ (95 % CI: 0.92–2.03) over 1925–2024. Galway also exhibits a relatively high mean rate of 1.38 mm yr⁻¹ (95 % CI: 0.90–1.98), while Cork shows a slightly lower rate of 1.28 mm yr⁻¹ (95 % CI: 0.80–1.85). These higher rates likely reflect the combined influence of the regional sea-level signal and spatial variations in vertical land motion, with weaker GIA-related uplift and relative subsidence contributing to enhanced RSL change in parts of southern and western Ireland. Tarbert exhibits the highest long-term rate, indicating the highest long-term RSL rate within the reconstructed network. Uncertainty is also greater at Cork, Galway, and Tarbert during the earlier part of the extended prediction period, with wider CIs also evident at Malin Head (Fig. 9). The increased uncertainty reflects reduced observational constraints during these periods, where the model relied more heavily on extrapolation beyond the available tide-gauge records. The extended prediction grid enables consistent rate estimates over 1925–2024, with greater uncertainty at sites having shorter observational records.

Comparison with the earlier period (1925–2000) highlights the recent acceleration in modelled RSL change. During 1925–2000, rates were generally low, with mean estimates ranging from 0.10 mm yr⁻¹ (95 % CI: -0.26–0.44) at Belfast to 0.83 mm yr⁻¹ (95 % CI: 0.21–1.64) at Tarbert. Malin Head, Dublin Port, Cork, and Galway showed rates of 0.65 (0.17–1.18), 0.49 (0.16–0.82), 0.67 (0.08–1.39), and 0.77 mm yr⁻¹ (0.19–1.54), respectively. Across individual tide-gauge sites, the modelled rates shown in Fig. 9 indicate higher rates during 2000–2024 than during the earlier period, consistent with a recent acceleration in RSL change across the tide-gauge network.

The regional modelled RSL signal is consistent with the acceleration evident in the tide-gauge observations (Fig. 10). The mean modelled RSL rate across the Irish tide-gauge network over 1925–2024 is 1.21 mm yr⁻¹ (95 % CI: 0.90–1.55). When separated into earlier and more recent periods, the mean rate increases from 0.59 mm yr⁻¹ (95 % CI: 0.19–1.02) during 1925–2000 to 3.13 mm yr⁻¹ (95 % CI: 2.45–3.81) during 2000–2024, indicating a marked increase in the rate of RSL change. The instantaneous mean modelled RSL rate in 2024 reaches 4.26 mm yr⁻¹ (95 % CI: 2.68–5.84), indicating that the rate of RSL change remained elevated in 2024. These mean modelled RSL rates incorporate the common regional sea-level signal together with site-specific GIA and local components, with GIA represented using the ICE6G_BC_hybrid_71p320 model.

The regional component exhibits a similar temporal pattern. Across the full model period (1925–2024), the posterior mean regional rate is 1.13 mm yr⁻¹ (95 % CI: 0.36–1.98). The regional rate increases from 0.51 mm yr⁻¹ (95 % CI: -0.34–1.45) during 1925–2000 to 3.04 mm yr⁻¹ (95 % CI: 2.06–3.94) during 2000–2024, reaching an instantaneous regional rate of 4.18 mm yr⁻¹ (95 % CI: 2.48–5.82) by 2024. The close correspondence between the mean modelled RSL rate and regional rate indicates that the common regional signal dominates the network-mean RSL evolution. The GIA-related and local components make comparatively small contributions to the network-mean rate, with the mean modelled RSL rate differing from the regional rate by < 0.10 mm yr⁻¹.

The accumulated RSL change was calculated from the difference between the posterior total modelled RSL estimates at the beginning and end of the reconstruction period. The mean accumulated RSL rise across the six reconstructed Irish tide-gauge sites was 11.8 cm (95 % CI: 8.78–15.1 cm) between 1925 and 2024.

This study presents an integrated view of RSL change for the island of Ireland over the past century. By incorporating newly digitized records from the south and southwest of the country, the analysis provides estimates of both the common regional sea-level signal and site-specific deviations from this signal. We find a consistent pattern of RSL rise, with acceleration becoming particularly pronounced since the early twenty-first century, with instantaneous RSL rates exceeding 4 mm yr⁻¹ at five of the six Irish tide-gauge sites in 2024. These recent rates indicate substantial acceleration in RSL rise around Ireland and are broadly consistent with the recent acceleration observed in global sea-level rise.

Results from the statistical modelling in this study are consistent with previous studies. Orford et al. (2006) reported RSL rates of -0.20 ± 0.37 mm yr⁻¹ for Belfast (1918–2002) and -0.16 ± 0.17 mm yr⁻¹ for Malin Head (1958–1998). For comparison, over the corresponding periods, the NI-GAM reconstruction estimates a modelled RSL rate of 0.10 mm yr⁻¹ (95 % CI: -0.26–0.44) for Belfast over 1925–2000 and 0.72 mm yr⁻¹ (95 % CI: 0.31–1.15) for Malin Head over 1958–1998. The Belfast estimate is consistent with the near-zero rate reported by Orford et al. (2006), while the Malin Head estimate is higher than the earlier estimate, although differences in record composition and methodological treatment limit direct comparison. These differences likely reflect variations in the underlying observations, record composition, temporal coverage, and methodological treatment of long-term trends. In particular, the updated Malin Head record differs in composition from that used by Orford et al. (2006), incorporating observations from Portrush in 1998. Differences in the available observations and record processing may therefore influence the estimated trend over this multi-decadal interval. The long-term mean rate at Malin Head is higher than at Belfast and broadly comparable to the rates observed at the southern and western sites, indicating that Malin Head exhibits relatively high RSL rates.

To the east, Dublin Port exhibits a long-term mean modelled RSL rate of 1.12 mm yr⁻¹ (1925–2024; 95 % CI: 0.83–1.41). The estimated mean modelled RSL rate for 1925–2000 was lower at 0.49 mm yr⁻¹ (95 % CI: 0.16–0.82), increasing to 2.69 mm yr⁻¹ (95 % CI: 2.16–3.23) for 2000–2016. The instantaneous RSL rate in 2024 reached 4.24 mm yr⁻¹ (95 % CI: 2.65–5.86), indicating a substantial recent increase in the rate of RSL change. These results are generally consistent with previous tide-gauge estimates for overlapping periods. Shoari Nejad et al. (2022) reported a tide-gauge-derived MSL trend of 1.1 mm yr⁻¹ (95 % CI: 0.6–1.6 mm yr⁻¹) for Dublin Port during 1953–2016 using Bayesian linear regression applied to corrected observations, compared with our modelled RSL rate of 1.08 mm yr⁻¹ (95 % CI: 0.83–1.32) for the same period. The two estimates are very similar and show broad agreement, providing consistent evidence of the long-term rate of RSL change at Dublin Port. At Dublin Port, our reconstruction indicates relatively modest and variable rates of RSL change prior to 2000, with modelled rates of -0.08 mm yr⁻¹ (1938–1961; 95 % CI: -0.57 to 0.39 mm yr⁻¹) and 0.08 mm yr⁻¹ (1961–1980; 95 % CI: -0.33 to 0.48 mm yr⁻¹). These estimates can be compared with historical values reported by Carter (1982), who estimated an MSL rate of 0.6 mm yr⁻¹ for 1938–1961 and an MSL rate of −0.3 mm yr⁻¹ for 1961–1980. Over 1938–1996, Woodworth et al. (1999) reported a long-term trend of 0.23 mm yr⁻¹, compared with our modelled RSL rate of 0.30 mm yr⁻¹ (95 % CI: 0.03–0.57). Differences between studies likely reflect variations in data coverage, corrections, and modelling approaches; however, all estimates indicate relatively modest rates of RSL change before 2000, followed by a pronounced acceleration in recent decades. Palaeo-reconstructions using saltmarsh foraminifera also indicate a comparable long-term RSL rate since the early nineteenth century (1.6 ± 0.9 mm yr⁻¹; Roseby et al., 2023).

Higher RSL rates of rise are observed along the southern and western coasts of Ireland. Cork shows a long-term mean modelled RSL rate of 1.28 mm yr⁻¹ (95 % CI: 0.80–1.85; 1925–2024), with the instantaneous RSL rate reaching 4.28 mm yr⁻¹ (95 % CI: 2.66–5.88) in 2024. This differs from the MSL rate reported by Pugh et al. (2021), who estimated a rate of 2.2 ± 0.2 mm yr⁻¹ for Cork over 1842–2019/2020. Differences between these estimates likely reflect contrasting time periods, modelling approaches, record processing, and differences in datum alignment. Pugh et al. (2021) aligned the Cork record to a datum of 2.752 m, whereas in this study the Cork reconstruction was aligned to the OPW datum of 2.69 m, with additional site-specific adjustments applied. Applying the 2.752 m datum results in a larger divergence between the Cork tide-gauge and satellite SSH trends, with a trend difference of approximately 2 mm yr⁻¹ in magnitude (tide gauge minus satellite). In contrast, the OPW-sourced datum provides closer agreement with the satellite and modern Ringaskiddy and Currach Club observations, providing further support for the adopted datum correction. The selected adjustment produced close agreement with modern Ringaskiddy and Currach Club observations from 2012 onwards, providing support for the adopted datum correction. Tarbert exhibits the highest long-term mean rate among the analysed sites, with a mean modelled RSL rate of 1.43 mm yr⁻¹ (95 % CI: 0.92–2.03; 1925–2024). This rate is consistent with the broader pattern of higher RSL rise along southern and western Ireland but should be interpreted with some caution because the historical Tarbert record has greater datum uncertainty than other sites. The instantaneous RSL rate in 2024 reached 4.41 mm yr⁻¹ (95 % CI: 2.75–6.02). River discharge from the River Shannon may also influence observed sea-level variability at Tarbert, as shown in other estuarine systems (e.g., Piecuch et al., 2018). On the west coast, Galway exhibits a long-term mean modelled RSL rate of 1.38 mm yr⁻¹ (95 % CI: 0.90–1.98; 1925–2024), with the instantaneous RSL rate increasing to 4.37 mm yr⁻¹ (95 % CI: 2.73–5.99) in 2024, consistent with the broader regional acceleration in RSL change.

The reconstructed Irish RSL record (1925–2024) was compared with a global tide-gauge MSL reconstruction and a global satellite-altimetry SSH record, using the Palmer et al. (2021) global tide-gauge reconstruction (1925–2010) and the Copernicus Climate Change Service (C3S, 2018) satellite-altimetry sea-surface height (SSH) record (1993–2024) (Fig. 11). All three series were referenced to 1993 (1993=0), providing a common baseline for comparison of their temporal evolution. Prior to the late twentieth century, the Irish and global tide-gauge series showed different temporal behaviour. While the global tide-gauge reconstruction exhibited an overall long-term rise, the Irish modelled RSL increased only modestly during the mid-twentieth century, rising by approximately 1.3 cm between 1925 and 1975. The two records became more closely aligned during the latter part of the twentieth century. From approximately the 1990s to 2010, the Irish modelled RSL followed a broadly similar upward trajectory to the global tide-gauge reconstruction, although the global series exhibited greater interannual variability and differences in magnitude occurred at various times.

During the satellite era (1993–2024), the Irish modelled RSL and global satellite-altimetry SSH record showed broadly similar upward trajectories. Both indicated substantial sea-level rise over this period, although the global satellite series rose somewhat more strongly, exceeding 0.10 m above the 1993 reference by 2024, compared with approximately 0.09 m for the Irish modelled RSL. The broadly similar late-period trajectories demonstrate that both the Irish reconstruction and the global satellite-altimetry record exhibit a pronounced rise in sea level during the satellite era.

Between 1925 and 2024, the mean modelled RSL rate across the Irish tide-gauge network was 1.21 mm yr⁻¹ (95 % CI: 0.90–1.55 mm yr⁻¹). The model indicates a clear acceleration through time, with the mean modelled RSL rate increasing from 0.59 mm yr⁻¹ (95 % CI: 0.19–1.02 mm yr⁻¹) during 1925–2000 to 3.13 mm yr⁻¹ (95 % CI: 2.45–3.81 mm yr⁻¹) during 2000–2024. The cumulative RSL change, calculated from the difference between posterior total modelled RSL estimates in 1925 and 2024, was 11.8 cm (95 % CI: 8.78–15.1 cm). This magnitude is broadly consistent with reported twentieth- and early twenty-first-century global MSL rise estimates of approximately 10–25 cm (IPCC, 2021; Mu et al., 2025), although regional differences arise from spatial variability in processes such as GIA, ocean dynamics, and local VLM.

Hogarth et al. (2021) reconstructed long-term sea-level records for Great Britain using an extensively augmented and spatially clustered tide-gauge dataset, applying GIA corrections, and reported a post-1900 weighted linear rate of 2.12 mm yr⁻¹, with evidence of long-term acceleration from the nineteenth to twentieth century. The mean modelled RSL rate estimated here across the Irish tide-gauge network was 1.21 mm yr⁻¹ (95 % CI: 0.90–1.55 mm yr⁻¹) over 1925–2024, while the corresponding regional RSL component was 1.13 mm yr⁻¹ (95 % CI: 0.36–1.98 mm yr⁻¹). The lower mean modelled RSL rate estimated here for Ireland may partly reflect differences in regional sea-level response and GIA-related VLM, while differences in tide-gauge spatial coverage, record length, temporal periods analysed, and modelling approaches limit direct comparison. Earlier tide-gauge studies, as noted previously, reported near-zero to slightly negative RSL trends in Northern Ireland during the late 1990s and early 2000s (Orford et al., 2006). The present results are broadly consistent with these findings and with wider evidence indicating relative uplift in Northern Ireland, while southwestern Ireland experiences reduced uplift or subsidence (Bradley et al., 2011; Bradley et al., 2023; Kirby et al., 2023). These spatial differences reflect variations in GIA associated with former ice loading, with greater ice thickness and loading over Britain producing stronger GIA-related VLM responses, while thinner ice cover over Ireland resulted in comparatively smaller GIA effects (Shennan et al., 2018).

The GIA model ICE6G_BC_hybrid_71p320 was selected because it produced the smallest inter-site spread in GIA-corrected RSL trends across the Irish tide-gauge network. However, this selection should not be interpreted as indicating that it represents the uniquely optimal geophysical reconstruction of GIA. Rather, it represents the most suitable correction according to the criterion adopted here, namely reducing spatial variability in observed RSL trends prior to integration into the reslr model. The inclusion of the modern Bangor record, adjacent to the Belfast tide-gauge record, may contribute to the recent increase in northern RSL rates and reduce the apparent difference between northern and southern sites. Data gaps and differences in record quality were carefully considered during data preparation. While the Bangor observations were independently inspected and quality-controlled, the BODC warning that observations from approximately 2016 onwards remained potentially suspect indicates that residual uncertainty remains in the later Bangor record, and its data quality cannot be fully assured. To avoid potentially influential recent observations, the 2023–2024 Bangor data were excluded from the analysis because the apparent acceleration was considered potentially unreliable. Inclusion of these observations would increase the estimated Belfast rate over 1925–2024 from the current estimate of 0.73 to ∼ 0.96 mm yr⁻¹. During the earlier period, Belfast showed a near-zero mean rate of 0.10 mm yr⁻¹ (95 % CI: -0.26–0.44) over 1925–2000, compared with 0.30 mm yr⁻¹ (95 % CI: -0.02–0.62) over 1925–2010. Both estimates indicate relatively limited long-term RSL rise, although the 1925–2010 estimate is higher and remains associated with substantial uncertainty. These earlier estimates indicate relatively low rates of RSL rise, with substantial uncertainty, whereas inclusion of the modern Bangor observations results in a more pronounced recent acceleration in the Belfast record. The sensitivity of the estimated rates to the choice of GIA model highlights the importance of model selection and indicates that differences between sites should be interpreted cautiously. In the present analysis, ICE-6G variants generally produced the smallest spatial gradients across the Irish network, while GLAC1D variants produced intermediate gradients. Bradley-type models generally produced the largest spatial gradients, resulting in a stronger north–southwest contrast in modelled RSL trends.

Cross-validation of the NI-GAM model indicates generally good predictive performance for the reconstructed RSL time series. Three-fold cross-validation showed that the 95 % prediction intervals had empirical coverage of 72.0 %–92.6 % across sites, with an overall coverage of 90.21 %, indicating some under-coverage relative to the nominal 95 % level. Prediction intervals were highly consistent across the network, with widths ranging from 0.122 to 0.125 m. The near-zero overall mean error indicates limited overall bias in the model predictions. Mean absolute error (MAE) and root mean squared error (RMSE) were low, with overall values of 0.028 and 0.036 m, respectively, while site-specific RMSE values ranged from 0.0321 to 0.0510 m. Some variation is evident among sites, with Tarbert showing the largest residual errors and lowest predictive coverage, which likely reflects greater uncertainty associated with the sparse and discontinuous historical record and limited observational constraints. Belfast and Dublin Port also showed some departures from the 1:1 observed–predicted relationship, potentially indicating that some observed multi-decadal variability was not fully reproduced by the model at these sites.

Although the tide-gauge records were represented as annual observations in the model inputs, a common observational uncertainty may not fully represent site- and time-dependent variability associated with the underlying measurements. Future work could investigate whether incorporating time-dependent and site-specific observational uncertainties, derived from the underlying tide-gauge records, improves uncertainty calibration and predictive agreement at locations with greater local variability. Overall, the cross-validation results indicate that the NI-GAM provides generally good predictive performance across the network, while highlighting locations where local variability and uncertainty representation may influence predictive performance.

Some uncertainty remains in the observational data, particularly related to historical datum conversions, metadata limitations, and the interpretation of incomplete historical documentation. Although extensive datum adjustments and sensitivity assessments were undertaken to standardise records to Ordnance Datum Malin, uncertainties in historical benchmarks, reference relationships, and undocumented changes in gauge configuration may influence long-term RSL estimates. Additional observational uncertainties include potential digitization errors in historical records (McLoughlin et al., 2026), instrumentation effects (Gobron et al., 2019), and residual high-frequency variability not fully removed by atmospheric corrections. For Belfast, additional uncertainty may arise from incomplete temporal coverage in the historical record digitized by Murdy et al. (2015). In some instances, observations were recorded or digitized a few minutes after the nominal hourly interval and were therefore not retained when the record was subsequently filtered to exact hourly timestamps. This can result in gaps in the standardized hourly series and may introduce uncertainty into daily, monthly, and annual mean estimates where temporal coverage is incomplete. Although atmospheric corrections accounted for pressure- and wind-driven variability, annual means based on incomplete records may not fully represent the seasonal cycle and could therefore introduce small positive or negative biases in individual years. The seasonal harmonic terms included in the regression remove the modelled seasonal component from the monthly records, but residual seasonal variability may remain where observational coverage is incomplete.

The reslr package produces robust regional estimates by integrating data from multiple sites (Upton et al., 2024). This hierarchical approach allows sites with shorter or more incomplete records, such as Cork, Tarbert and Galway, to borrow information from the regional signal and other sites included in the model, providing continuous estimates across the full prediction period when prediction grids are extended. The conservative spline specification adopted in the NI-GAM prioritises estimation of long-term regional variability while limiting the representation of shorter-term variability that is not well constrained by the available tide-gauge observations.

For sites with limited observations, the model combines site-specific information with the regional signal and the linear (GIA-related) component to estimate RSL changes beyond the observed record. Estimates within the observation period are therefore most strongly constrained, whereas extrapolated estimates rely more heavily on model assumptions and regional information and should be interpreted with greater caution.

The Irish tide-gauge network analysed in this study shows evidence of accelerating RSL rise, with spatial differences reflecting local processes and higher long-term rates along the southern and western coasts. These regions, including Cork, Tarbert, and Galway, may warrant particular attention because they exhibit relatively high modelled RSL rates and, in some cases, limited observational coverage, particularly within the Shannon Estuary. Projected long-term average rates towards the end of the century are highest in the southwest (3.3–4.8 mm yr⁻¹) and lower in the northeast (2.2–3.7 mm yr⁻¹) (MWP, 2024). Enhanced tide-gauge monitoring would improve understanding of local variability and support adaptation planning, particularly considering the projected acceleration in global MSL (Wang et al., 2024).

The key objective of this study was to estimate historical sea-level change across Ireland, with particular focus on the south and southwest, where long-term observations have been limited. By integrating archival and modern tide-gauge data, including Cork, Tarbert, and Galway, this study fills critical spatial gaps and situates Irish sea-level changes within a broader regional context, complementing previous studies in the UK.

In 2024, modelled instantaneous RSL rates were elevated across Ireland, ranging from 3.89 to 4.41 mm yr⁻¹. Rates were broadly similar among sites, with the highest estimated values occurring at Malin Head and Tarbert, followed by Galway, Cork, and Dublin Port, while Belfast exhibited the lowest estimated rate. These results indicate widespread acceleration in RSL rise across the Irish tide-gauge network, although the 2024 Belfast estimate reflects modelled continuation beyond the end of its observational record in 2015. Adjustments for datums, atmospheric effects, seasonal cycles, and GIA were incorporated into the analysis, with additional sensitivity testing used to assess the influence of GIA model choice. Under the ICE6G_BC_hybrid_71p320 model, the site-specific GIA rates were selected because this model produced the smallest inter-site spread in observed RSL trends when evaluated using a linear regression framework prior to integration into the reslr package. This selection does not indicate that it represents the definitive best GIA model. Future studies could further quantify the influence of GIA uncertainty and evaluate which GIA model provides the most physically plausible representation of VLM.

Cross-validation indicates generally good predictive performance of the regional reconstruction, with overall predictive coverage of 90.21 %, reasonably close to the nominal 95 % level, low overall error, and broadly consistent prediction-interval widths across the network. While short-term variability at individual sites is smoothed, the model captures the broad patterns of observed RSL variability and provides generally consistent uncertainty estimates. Performance varies among sites, with Tarbert showing the lowest predictive coverage (72.0 %) and highest prediction errors, which likely reflects greater uncertainty associated with its sparse and discontinuous observational record. Overall, this study demonstrates the value of integrating modern statistical frameworks with archival data to reconstruct historical RSL change across Ireland. By synthesising available MSL observations from across the island, this study provides the first geographically balanced reconstruction of historical sea-level change in Ireland and strengthens the foundation for scientific understanding and evidence-based policy development. Previous studies of MSL change in Ireland reported relatively low rates of rise (Carter, 1982; Woodworth et al., 1999; Orford et al., 2006), while more recent individual site studies have identified rates of sea-level rise consistent with climate change-driven trends (Shoari Nejad et al., 2022). This synthesis reveals a clear acceleration in RSL rise since the late twentieth century, with the increase becoming particularly pronounced during the early twenty-first century. The mean modelled RSL rate increased from 0.59 mm yr⁻¹ (95 % CI: 0.19–1.02) during 1925–2000 to 3.13 mm yr⁻¹ (95 % CI: 2.45–3.81) during 2000–2024. The results indicate that the recent acceleration in RSL rise around Ireland is broadly consistent with the acceleration observed in global sea-level rise, with higher long-term modelled RSL rates generally occurring at southern and western sites, while lower rates are generally observed at northern sites. Given the observed recent acceleration in sea-level rise and projected future increases, Ireland's coastal population will need to adapt to the associated increase in coastal flooding and inundation. Future work should focus on extending this reconstruction to additional sites and refining GIA corrections to improve understanding of regional sea-level change and support adaptation planning in vulnerable coastal areas.