Abstract Millions of people are vulnerable to future sea level rise (SLR); however, SLR predictions are less accurate than temperature and atmospheric CO2 concentration outcomes under different gas emission scenarios. There are also lags in the analytical determinations and uncertain future changes in climate forcing functions. A linear relationship exists between sea level elevation and atmospheric carbon dioxide concentration (mm ppm-1; SEACO) at 10 tide gage stations across the world of 0.005911 mm ppm-1 y-1 from 1901 to 2020. This ratio can serve as a diagnostic tool for SLR tipping points, help calibrate SLR models, and be easily calculated for regional oceans. Higher or lower future regional changes may indicate tipping points in ocean current strength or direction or foreshadow global climate changes. A regional example is from the Gulf of Mexico (aka Gulf of America), where, after 2000, SEACO y-1 was more than twice as high as in the previous six to eight decades. The predicted atmospheric CO2 concentrations in the high-end emissions IPCC scenario (RCP8.5) for 2050 and 2100 are 550 and 1000 ppm, respectively. If these concentrations occur and the SEACO index continues to rise at the same rate as during the last 120 years, then the average annual sea level rise (SLR) from 2020 to 2050 will be 164% of the 1991 to 2020 baseline years (2.17 to 3.57 mm y-1), and the 2020 to 2100 interval average will be 278% of the baseline years, equivalent to an annual SLR of 3.57 and 6.73 mm y-1, respectively. At those concentrations the average global sea level in 2050 and 2100 will be 0.11 and 0.54 m, respectively, above the 2020 average elevation and considerably below the sea level predicted under the IPCC RCP8.5 high-emissions scenario.

Analysis of the uncertainties in tidal constants obtained from short tide gauge records and their value for tidal studies

Understanding suspended sediment transport in macrotidal embayments is crucial for assessing water quality, ecosystem function, and long-term morphological stability. This study provides a high-resolution, localized estimate of suspended sediment flux and examines the empirical relationship between turbidity (NTU, nephelometric turbidity unit) and total suspended matter (TSM, mg·L−1) in the main tidal channel of Gomso Bay, a UNESCO-designated tidal flat on the west coast of Korea. A 13 h high-resolution fixed-point observation was conducted during a semi-diurnal tidal cycle using a multi-instrument platform, including an RCM, CTD profiler, tide gauge, and water sampling for gravimetric TSM analysis. Vertical measurements at the surface, mid, and bottom layers, taken every 15–30 min, revealed a strong linear correlation (R2 = 0.94) between turbidity and TSM, empirically validating the use of optical sensors for real-time sediment monitoring under the highly dynamic conditions of Korean west-coast tidal channels. The net suspended sediment transport load was estimated at approximately 5503 kg·m−1, with ebb-dominant residual currents indicating a net seaward sediment flux at the observation site. Residual flows over macrotidal channels are known to vary laterally, with landward fluxes often occurring over shoals. Importantly, the results from this single-station, short-duration observation indicate a predominantly seaward suspended sediment transport during the study period, which should be interpreted as a localized and time-specific estimate rather than a bay-wide characteristic. Nevertheless, these findings provide a baseline for assessing sediment flux and contribute to future applications in digital twin modeling and coastal management. Gomso Bay is part of the UNESCO-designated ‘Getbol, Korean Tidal Flats’, underscoring the global significance of preserving and monitoring this dynamic coastal system.

This study analyzes mean sea level variability in the Canary Islands from 1993 to 2022 using tide gauge and satellite altimetry data. During this period, both Las Palmas de Gran Canaria and Santa Cruz de Tenerife exhibited a significant sea level rise of 4.04 ± 0.83 and 4.38 ± 0.93 mm yr−1, respectively. Comparison between tide gauge and altimetry records reveals slight land subsidence at both locations, approximately 0.5–0.7 ± 0.55 mm yr−1, contributing to the observed relative sea level rise. The spatial differences in the trends observed from altimetry appear to be associated with mesoscale ocean dynamics, particularly an increase in eddy activity along the Canary Eddy Corridor. Projections based on IPCC SSP scenarios suggest that sea level could rise by up to 395 mm in Santa Cruz and 365 mm in Las Palmas by 2050 under high-emission conditions. An additional 20 mm could be added due to land subsidence if it remains constant. Interannual variability is primarily correlated with the North Atlantic Oscillation (NAO); however, Atlantic Multidecadal Oscillation (AMO) and the Atlantic Meridional Overturning Circulation (AMOC) indices also appear to correlate well with its low-frequency components. The seasonal cycle, driven primarily by steric effects, peaks in late summer and reaches a minimum in late winter, with its amplitude varying across the region. The seasonal amplitude is approximately 49.6 mm in Las Palmas and 70.2 mm in Santa Cruz.

Abstract. Peaks over threshold (POT) techniques are commonly used in practice to model tail behaviour of univariate variables. The resulting models can be used to aid in risk assessments, providing estimates of relevant quantities such as return levels and periods. An important consideration during such modelling procedures involves the choice of threshold; this selection represents a bias-variance trade-off and is fundamental for ensuring reliable model fits. Despite the crucial nature of this problem, most applications of the POT framework select the threshold in an arbitrary manner and do not consider the sensitivity of the model to this choice. Recent works have called for a more robust approach for selecting thresholds, and a small number of automated methods have been proposed. However, these methods come with limitations, and currently, there does not appear to be a “one size fits all” technique for threshold selection. In this work, we introduce a novel threshold selection approach that addresses some of the limitations of existing techniques, which we have termed the TAil-Informed threshoLd Selection (TAILS) method. In particular, our approach ensures that the fitted model captures the tail behaviour at the most extreme observations, at the cost of some additional uncertainty. We apply our method to a global data set of coastal observations, where we illustrate the robustness of our approach and compare it to an existing threshold selection technique and an arbitrary threshold choice. Our novel approach is shown to select thresholds that are greater than the existing technique. We assess the resulting model fits using a right-sided Anderson-Darling test, and find that our method outperforms the existing and arbitrary methods on average. We present and discuss, in the context of uncertainty, the results from two tide gauge records; Apalachicola, US, and Fishguard, UK. In conclusion, the novel method proposed in this study improves the estimation of the tail behaviour of observed coastal water levels, and we encourage researchers from other disciplines to experiment using this method with their own data sets.

Storm surges present a major hazard to coastal areas worldwide, a risk that is further amplified by ongoing sea-level rise associated with climate warming. The purpose of this study is to enhance the prediction performance of a storm surge height model by incorporating data resampling techniques into a multiple linear regression framework. Typhoon-related predictors, such as location and intensity-related parameters, were used to estimate observed storm surge heights at eleven tide gauge stations in southeastern Korea. To address the data imbalance inherent in storm surge height distributions, we applied combinations of over- and under-sampling methods across various threshold levels and evaluated them using four statistical metrics: root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), and the coefficient of determination (R2). The results demonstrate that both threshold selection and sampling configuration significantly influence model accuracy. In particular, station-specific sampling strategies improved R2 values by up to 0.46, even without modifying the regression model itself, underscoring the effectiveness of data-level balancing. These findings highlight that adaptive resampling strategies—tailored to local surge characteristics and data distribution—can serve as a powerful tool for improving regression-based coastal hazard prediction models.

This study reconstructs and projects relative sea-level change (RSLC) along Chesapeake Bay, a global hotspot for sea-level rise, from 1900 to 2100 by statistically extrapolating observed tide gauge trends, rather than employing climate model-based scenarios. The approach integrates global mean sea-level rise (GMSLR), regional sea-level rise (RSLR), and local land subsidence (LS) to evaluate both past and future behavior. Tide gauge data reveal that Chesapeake Bay’s sea level has accelerated at 0.099 ± 0.013 mm/year2 since 1992, with a linear rate of 1.26 mm/year since 1900, slightly outpacing global averages. LS, primarily driven by glacial isostatic adjustment (GIA) and sediment compaction, has been the dominant contributor to RSLC since the early 20th century, accounting for up to 71% of the RSLC prior to 1992 across 15 tide gauge stations. However, with GMSLR accelerating at 0.120 ± 0.025 mm/year2, the relative contribution of LS to RSLC is projected to decline to 31–43% by 2100. The reconstructed RSLC for the 20th century ranges between 32 and 44 cm, while extrapolated projections for the 21st century indicate a further increase of 53–99 cm. By 2100, GMSLR is expected to contribute to 60–70% of total RSLC. Spatial variability in RSLC across 15 tide gauge stations reflects differing geological conditions and anthropogenic influences such as groundwater withdrawal and construction-induced subsidence. These findings highlight the critical need for adaptive strategies to mitigate the impact of rising sea levels on coastal communities and infrastructure in the Chesapeake Bay region. Continued monitoring, improved modeling, and targeted resilience planning are essential to address the accelerating threats posed by sea-level rise and to ensure the sustainability of vulnerable coastal areas.

ABSTRACT While long-period signals in sea-level records are increasingly recognized as critical components of coastal dynamics, their precise reconstruction, quantification, and geophysical attribution remain elusive. This study introduces an extended multivariate Hankel spectral analysis (MHSA) framework that incorporates a stacked Hankel z-pole algorithm, achieving the enhanced spectral resolution and precision signal decomposition for the detection of long-term sea-level oscillation. By applying this methodology to global tide gauge {TG} and global mean sea-level (GMSL) records, we resolve a persistent 64.2 ± 0.8-yr oscillation (low-degree spherical harmonic (SH) mode, the variance explained >30% of long-term residuals) that may differ from previously documented periods and phases of climate-driven datasets. Three findings emerge: (1) A common variable mode detector specifying this signal’s prominent spatiotemporal coherence (strip-shaped SH distribution) across most coastal tide gauges is developed. (2) Monte Carlo simulation reveals that this SH-mode oscillation induces non-negligible deviations in sea-level trajectory models. For example, it minimally affects linear trends (2.1 ± 0.3 mm/yr) but can amplify acceleration budgets by >50% in short-scale (e.g. sub-cyclical records) tide gauge datasets. Furthermore, a 17.8 ± 2.3 mm in recent 30-yr altimetry sea-level projections is introduced. (3) MHSA approach reveals sea-level variations waveform- and phase-locked with the Earth’s rotation and magnetic field while asynchronous with possible climate-driven changes and activity. These results (SH distribution and waveform consistency of geophysical processes) imply an internal (core–mantle) origin rather than climatic forcing. Therefore, the MHSA methodology offers an effective tool for exploring long-period sea-level variations, which is benefit for the future sea-level projections and coastal-risk evaluation.

Some intertidal corals, known as microatolls, have a distinct morphology that reflects changes in local relative sea level. While past observations have shown that the top surface of these corals may be killed by subaerial exposure, little is known about the exact oceanographic or environmental conditions that cause a coral to die down to a particular level. Here, we combine field surveys, tide-gauge data and analysis of microatoll morphology to investigate the survival limits of Porites spp. microatolls on Singapore’s intertidal reefs. Unponded Porites spp. microatolls on the Pulau Biola reef reach a ‘highest level of growth’ between mean low water springs and mean low water neaps. Diedowns on the highest microatolls during 2023 and 2024 suggest the survival of these corals depends on the duration of subaerial exposure. By comparing the estimated highest level of survival after a diedown to water levels recorded at local tide gauges, we show that intertidal corals on the Biola reef and nearby Siloso Point reef can survive more than 2 h of continuous subaerial exposure per day. However, Porites spp. corals may not have survived more than 3.5 h of daily partial exposure without dying down.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-22805-6.

The Chesapeake Bay (CB) region faces significant risks from relative sea-level change (RSLC), driven by global mean sea-level rise (GMSLR), regional sea-level rise (RSLR), and local land subsidence (LS). This study introduces a methodology to decipher RSLC trends in the CB area by integrating these components. We develop trend equations spanning 1900–2100, incorporating acceleration for GMSLR and RSLR since 1992, with linear LS estimation using tide gauge, satellite altimetry, and InSAR data. Our approach employs dynamic RSLC equations, Maclaurin series expansions, and inverse simulations to project RSLC trends through 2100. Stable RSLC rates require over 122 years of data for reliable linear trend estimation, with the Baltimore tide gauge providing the necessary long-term dataset. Similarity in monthly mean sea-level variations within a coastal region enables a new method to identify LS from short-term tide gauge data by correlating it with corresponding long-term data at Baltimore. LS is categorized into bedrock-surface subsidence (BSS) and compaction subsidence (CS), with methods proposed to map BSS contours and estimate CS. CS is further classified into primary consolidation, secondary consolidation, construction-induced, and negative subsidence to determine specific compaction types. The projection model highlights the dominant influence of GMSLR acceleration since 1992, with local LS and RSLR influenced by ocean circulation, density changes, and gravitational, rotational, and deformational (GRD) effects. This integrated approach enhances understanding and predictive reliability for RSLC trends, supporting resilience planning and infrastructure adaptation in coastal CB communities.

Abstract. Repeated absolute gravity measurements, conducted once or twice per year, have proven valuable for quantifying slow vertical land motion with a precision better than 0.4 µGal yr−1 (1 µGal = 10−8 m s−2) after a decade or more. This precision is comparable to vertical velocity estimates derived from continuously operating space-based geodetic techniques such as the Global Navigation Satellite System (GNSS). Furthermore, absolute gravimeters are particularly well suited for long-term studies, as their measurements are based on fundamental length and time standards (laser and atomic clock) and remain independent of terrestrial reference frame realizations, unlike GNSS. Consequently, an absolute gravimeter can return years or even decades later and provide relevant measurements, provided the initial gravity data are well documented and the ground gravity marker remains undisturbed. Following this line of thinking, we have compiled and consistently reprocessed absolute gravity measurements collected between 1998 and 2022 in Brest, on the French Atlantic coast, near its century-long tide gauge station. The entire dataset has been reanalyzed in accordance with international recognized standards for instrumental and modelling corrections. This effort has yielded a 25-year time series of absolute gravity values, which we present and document for future studies, along with details on our reprocessing methodology. We assess the quality of this dataset and evaluate the extent to which the observed linear gravity trend agrees with vertical velocity estimates from the nearby GNSS station co-located with the tide gauge. The gravity data and metadata are made available via the French hydrographic agency Shom portal (https://doi.org/10.17183/DATASET_GRAVI_BREST; Lalancette et al., 2024).

Climate-driven sea level rise (SLR) will intensify extreme sea level (ESL) events along China’s coast. This study reconstructs continuous hourly total sea level (TSL) by incorporating SLA, tide, storm surge, and wave components, addressing the sparse coverage of tide gauges and their often overlooked extreme events (waves). Using a peak-over-threshold method (thresholds optimised between the 97.00-99.99th percentiles) and a 3-day declustering interval, extreme samples were fitted with generalised Pareto distributions via maximum likelihood estimation. Present-day 1-year return levels exceed 1.6 m at most stations, while centennial return levels surpass 4.1 m at Lusi and Kanmen. Spatial variability is evident, with the East China Sea exhibiting higher extremes (~ 3.9–4.7 m for 100-year events) compared to the Yellow Sea and Bohai Sea (~ 2.7–3.5 m) and the South China Sea (~ 1.7–3.7 m). Under the SSP5-8.5 scenario, centennial return levels rise by 0.83 m by 2100, shifting 100-year events to below 50-year return periods by mid-century and less than 10 years by 2100. This study highlights the urgent need for regional adaptation strategies due to the rising intensity and frequency of extreme events, offering an approach that can be applied to other coastal systems to assess current and future ESL hazards under climate change.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-22493-2.

Storm surge dynamics in coastal zones and estuaries are complex, driven by coupled oceanic and terrestrial interactions that enhance the risk of coastal disasters. This study investigates storm surge characteristics and mechanisms in the Macao Cross Tidal Channel (MCTC), located in the Macao Sea Area (MSA). A tide-surge coupled numerical model was established using the unstructured grid Finite Volume Community Ocean Model (FVCOM). The model was rigorously validated against tide gauge data from Typhoon Hato, demonstrating strong performance, with a skill score of 0.95 and a correlation coefficient exceeding 0.94. The spatiotemporal characteristics and mechanisms of storm surge dynamics in the MCTC were elucidated. The results show that the MCTC’s complex geometry induces a geometric funneling effect, which substantially amplifies the storm surge compared with adjacent locations in the estuary and open sea. During the typhoon period, coastal geomorphology affects winds, tide levels, currents, and waves, which in turn spatially and temporally modulate the storm surge. Wind is the primary driver, but its effect is modulated by nonlinear interactions with waves, including enhanced bottom friction and wave set-down. In isolation, the wind-induced component contributed approximately 106% of the peak total surge. This overestimation quantitatively highlights the critical role of nonlinear interactions, where wave-enhanced bottom friction acts as a major energy sink, and wave set-down directly suppresses the water level at the channel entrance. The individual peak contributions from atmospheric pressure and wave were approximately 5% and 17%, respectively, but these peaks occurred out of phase with the storm surge. Energy transformation analysis based on the Bernoulli principle revealed a distinct conversion from potential to kinetic energy in the constricted transverse waterway, while the longitudinal waterway exhibited a more gradual energy change. These findings enhance the mechanistic understanding of storm surges in complex, constricted estuaries and can inform targeted strategies for coastal hazard mitigation in the Macao region.

Coastal flooding poses a significant risk to Indonesia’s low-lying coastal zones, where sea level rise and extreme events intensify vulnerability. This study explores the integration of Jason-3 satellite altimetry and tide gauge observations in Banten Bay to enhance coastal flood prediction. The methodology includes datum harmonization, aliasing analysis, and harmonic reduction to refine sea level estimates. Results demonstrate that the synergy between altimetry and tide gauge data reduces residual errors, improves the identification of non-tidal variability, and strengthens the accuracy of flood predictions compared to using individual datasets. The findings highlight the potential of this combined approach to support coastal monitoring and early warning systems in regions with limited observational networks. While limitations remain, including reliance on a single tide gauge and uncertainties in geophysical corrections, the framework is adaptable to other Indonesian coastal areas and scalable to regional and global applications in disaster risk reduction.

Abstract In storm surge (SS) simulation, data‐driven methods can establish the relationship between predictor variables and the predictand, enabling long‐term SS level reconstructions. Here, using the U.S. East Coast as an example, we explored the capabilities of four machine learning algorithms, namely Artificial Neural Networks (ANN), Long Short‐Term Memory (LSTM), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost) in reconstructing hourly SS levels from 1979 to 2018 under an all‐site modeling framework. Four atmospheric parameters, time index, and tide gauge coordinates from 51 tide gauges are used as predictors. The model performance was evaluated at both the tide gauge and coastal scales. Results indicate that LightGBM and XGBoost models outperform ANN and LSTM in SS reconstructions, with XGBoost showing better overall performance, especially for extreme SSs and historical extreme events. XGBoost can capture the temporal evolution of SSs with higher accuracy, producing reconstructions comparable to observations under the all‐site modeling framework. The model interpretability analysis focusing on XGBoost reveals that the spatial distribution of feature importance varies for each predictor. Mean sea level pressure and the 10 m eastward wind component are the two most important predictors, followed by time index, latitude, and longitude under the all‐site modeling framework and selected stations. These results indicate that data‐driven models under this framework have the potential to capture region‐specific and physically reasonable relationships between SS levels and atmospheric drivers.

Abstract The on-shelf penetration of low-frequency open-ocean signals makes a significant contribution to the variability of coastal sea level. However, owing to the complicated coupling, the high-frequency tidal effects on the shoreward penetration of the low-frequency signals are generally overlooked. This study revisits the classic β-plane arrested topographic wave model aiming to more explicitly reveal the role of tides in modulating the open-ocean sea level transmission over the continental shelf. By inferring and comparing different forms of the vorticity equation, we reexpress the bottom friction coefficient (BFC) as based on a linear eddy-viscosity parameterization, thereby relating BFC to the thickness of the bottom boundary layer δB and further to the bed shear stress (BSS) τb (f being the Coriolis parameter, κ being a constant, and ρ is the seawater density). This provides a novel perspective to examine tidal effects on the across-shelf transmission by estimating enhanced BSS induced with the addition of tidal currents. Using appropriate parameterizations to estimate BSS, we apply the calculations to the western North Atlantic. It is shown that BFC exhibits an abrupt increase between 28° and 35°N by including tidal currents, which enhances the on-shelf penetration of open-ocean signals, especially in the downstream vicinity of 31°N. Moreover, modeling experiments indicate that this enhancement is more evident for shorter-wavelength signals. Such a pronounced coastal response is clearly manifested in tide gauge measurements along the east coast of North America. We also discuss the impact of tidal current rotation on the ocean-to-coast transmission for a constant eddy-viscosity scenario. Significance Statement Large-scale, low-frequency sea level signals from the deep ocean propagate across the continental slope and shelf to the coast, significantly altering coastal sea level variations. However, the effect of high-frequency tides on the shoreward transmission of open-ocean signals is largely overlooked. We rederive a new expression for the bottom friction coefficient, incorporating the seabed shear stress, allowing us to more explicitly examine the tidal effects. The modeling solutions show that the tidal currents aid the cross-shore transmission particularly near 31°N, and the coast is more likely to experience shorter-wavelength signals from the deep ocean due to the addition of tidal currents. Such an enhanced shoreward transmission near 31°N is clearly observed as a pronounced coastal response from the tide gauge measurements along the east coast of North America.

Abstract. Coastally trapped waves (CTWs) are a major cause of sub-seasonal coastal sea level variability. While they have mostly been studied using numerical models, observational evidence is limited due to the sparse spatial coverage of the tide gauge network and the limitations of satellite altimetry gridded maps, which arise from the interpolation of sparse along-track data. The simultaneous operation of multiple altimetry missions, advancements in processing technologies, the advent of wide-swath altimetry, and the development of new interpolation techniques have the potential to significantly improve the monitoring of CTWs. In this study, we analyze three months of sea level data from satellite altimetry to evaluate the new capabilities for detecting sub-monthly variability, comparing the results to tide gauge data and an ocean model in Eastern Australia, an area known for its dominance of CTWs at these time scales. The results demonstrate that in the study area, the correlation between tide gauges and coastal daily sea level grids from satellite altimetry exceeds 0.5, even when considering time series filtered to capture only sub-monthly variability. CTWs are generally well detected, though discrepancies remain, particularly in terms of amplitude, wavelength and period.

Abstract. Saltwater intrusion is an increasing concern for coastal ecosystems. While groundwater models have made progress in simulating aquifer salinization, their boundary conditions – potentially informed by ocean model simulations in shallow water systems and intertidal zones – remain constrained. Here we presented a 3D unstructured-grid model that covers the Gulf of Maine and the Mid-Atlantic Bight, and most areas of the South-Atlantic Bight along the North American Atlantic Coast (“NAAC”) for 2 decades, with a focus on the salinity simulations. This model resolves detailed geometric features of tidal tributaries down to 100 m while maintaining a resolution of 6.5 km in the coastal ocean. The two-decadal simulations from 2001 to 2020 were evaluated using a comprehensive observational dataset of elevation, temperature, and salinity. The mean absolute error in the M2 amplitude across the NOAA tidal gauges within the domain is 0.11 m. The root-mean-square deviation for salinity and temperature measurements are 0.27 PSU and 0.12 °C, respectively. The model reasonably captured the currents and circulations. For the first time, we extended a regional continental scale ocean model to the tidal wetlands to include compound flooding process. The two-decade of simulations of hydrodynamic and hydrological connectivity along the Atlantic Coast have significantly addressed numerous observational gaps in many systems. Specifically, saltwater intrusion patterns in major estuaries of the Mid-Atlantic, such as Chesapeake Bay, Delaware Bay, and other tributaries within the same hydrologic unit, exhibit significant correlations. The seamless cross-scale capability of this model facilitates future applications to land-sea interactions, such as carbon fluxes.

Abstract. Tide gauges record sea level changes along coastlines. They are widely used to determine the twentieth century global mean sea level (GMSL) rise. However, a major issue in tide gauge data is the presence of various, substantial, and sometimes persistent data gaps, which hinder our understanding of sea level rise, especially at regional and local scales. Whilst the GMSL reconstructions have been provided by several influential studies, reconstructions at the exact sites of tide gauges are rarely available. Here, we present sea level reconstructions at global 945 tide gauges, covering the period from 1900 to 2022. Our approach relies on a data assimilation technique that integrates various physical sea level observations and predictions, including sea level simulations from 35 climate models. A prominent feature in our reconstruction is that it provides an ensemble of 35 reconstructions at each site of tide gauge, providing continuous and refined sea level time series. This ensemble reconstruction allows for direct statistical assessments, e.g., average, median, spread, and percentile. The average of reconstructed sea level across 945 tide gauges reveals a GMSL rate of 1.75 ± 0.05 mm yr−1 over 1900–2020, and shows strong agreement with other GMSL reconstructions for both the curves of time series and overall trends. At local scale, our reconstructions are comparable to an independent reconstruction. Despite some rate differences at certain locations, the reconstructed sea level trends closely follow the raw records when they are available, emphasizing the importance of the observations at tide gauges. Our sea level reconstructions offer a valuable resource for improving global and regional sea level projections, validating climate model performance, and informing coastal adaptation strategies through understanding the sea level rise over the past century. The reconstructed sea level is available at https://doi.org/10.5281/zenodo.15385035 (Mu, 2025).

Sea level variability in the East Sea (Sea of Japan) and the Northwest Pacific poses challenges for coastal risk management due to the scarcity of long-term observations at remote locations such as Dokdo (Dok Island). This study reconstructs a continuous monthly sea level record at Dokdo from 1993 to 2023 by imputing gaps in 13 nearby Permanent Service for Mean Sea Level tide gauge stations using eight machine learning models and geospatial interpolation methods. The ensemble mean of Machine Learning-based imputations produced physically realistic and temporally coherent timeseries, preserving both seasonal and interannual variability. Sea level at Dokdo, estimated via inverse distance weighting, aligned well with satellite altimetry from Copernicus Marine Service and exhibited strong regional coherence with nearby stations. These results demonstrate that a hybrid framework combining statistical imputation, Machine Learning, and inverse barometric correction can effectively reconstruct sea level in data-sparse marine regions. The methodology provides a scalable tool for monitoring long-term trends and validating satellite and model products in marginal seas like the East Sea.