Sea level variations have been continuously measured by satellite altimetry since 1993. In this study, we use the daily 0.25°×0.25° gridded sea level anomaly data, version DT2021 available from the Copernicus Climate Change Service (C3S) (Copernicus Climate Change Service, Climate Data Store, 2018; last access: 15 November 2025).

The data set is further corrected for the TOPEX-A instrumental drift that affected the first 6 years of the time series (Ablain et al., 2017), but the correction has no impact on our assessment that starts in 2004. The Jason-3 radiometer drift that impacts the wet troposphere correction (Brown et al., 2023) is corrected for. The altimetry data set is also corrected for the absolute Glacial Isostatic adjustment (GIA) effect, using the ICE6G-D model from Peltier et al. (2018).

The Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GRACE-FO) are joint missions by the National Aeronautics and Space Administration (NASA) and the German Aerospace Center (DLR) (Tapley et al., 2019). These satellites have been measuring temporal variations in Earth's gravity field since 2002. This data set is essential for estimating mass redistribution in the oceans, terrestrial water storage, ice sheets, and glaciers. GRACE/GRACE-FO data are generally available in two forms: Spherical Harmonic (SH) coefficients and Mass Concentration (mascon) solutions. In Bouih et al. (2025), we used, in addition to the mascon solutions, an ensemble of 60 SH solutions (update of Blazquez et al., 2018). The sea level budget residuals based on this ensemble mean SH solution showed a strong north–south signal in the mid-North Atlantic, due to north–south stripe noise affecting SH solutions (see Blazquez et al., 2018 for a discussion). For that reason, we focus here on the mascon solutions. In this study, we utilize the latest Release 6 (RL06) mascon solutions provided by three different institutions: the Center for Space Research (CSR, Save et al., 2016), the Jet Propulsion Laboratory (JPL, Watkins et al., 2015), and the Goddard Space Flight Center (GSFC, Loomis et al., 2019).

These mascon solutions, at monthly temporal resolution, are corrected for the geocentric motion (degree-1) using the Sun et al. (2016) solution. The C₂₀ and C₃₀ coefficients are derived from Satellite Laser Ranging (SLR) data. The GIA correction for GRACE is based on the ICE6G-D model from (Peltier et al., 2018).

Additionally, the GAD product derived from AOD1B models (Dobslaw et al., 2017; Flechtner et al., 2014), which represents non-tidal atmospheric and oceanic mass redistribution, is added back over the ocean areas to restore the ocean bottom pressure, which combines the effects of ocean mass and atmospheric loading. However, to align GRACE data with altimetry-based sea level (which is corrected for the inverse barometer effect), we convert ocean bottom pressure into manometric ocean mass by subtracting the monthly global ocean-mean atmospheric mass from every grid cell, using the spatial mean of the AOD1B RL06-based GAD product corresponding to each mascon solution (Chen et al., 2019). Note that using the AOD1B RL06 GAD product rather than any updated version is mandatory to maintain strict internal consistency with the background de-aliasing model applied during the RL06 mascon processing.

The CSR, JPL and GSFC products are provided with 0.25, 0.5 and 0.5° resolution, respectively. They are further interpolated on 1°×1° grids.

Finally, we use the ensemble mean of the three mascon solutions (CSR, JPL, GSFC) as our primary manometric product, deriving structural uncertainties from their ensemble dispersion across the basin (Blazquez et al., 2018; WCRP Global Sea Level Budget Group, 2018).

The Argo program is an international observational network that deploys a global array of autonomous profiling floats to measure temperature and salinity in the upper 2000 m of the ocean. Several institutions process the raw data from these floats and publish gridded temperature (T) and salinity (S) data sets. In this study, we utilize three distinct gridded products: the Scripps Institution of Oceanography (SIO, Roemmich and Gilson, 2009), the Japan Agency for Marine-Earth Science and Technology (JAMSTEC, Hosoda, 2007), and the Met Office Hadley Centre (EN4 product, version 2.2, Cheng et al., 2014). Note that the SIO product rejects salinity profiles exceeding a difference of 0.1 psu when comparing to historical estimate based on the WOCE Global Hydrographic Climatology (Roemmich and Gilson, 2009). This correction tends to remove the salinity drift reported in Argo floats since 2016 (Liu et al., 2020; Ponte et al., 2021; Wong et al., 2023). The other two products do not apply such a correction. All three data products are available at https://argo.ucsd.edu/data/argo-data-products (last access: November 2025). The gridded Argo-based time series have a spatial resolution of 1°×1° at monthly interval over January 2004 to December 2022.

We also used the ocean reanalysis data set CNR-ISMAR Global Historical Reanalysis (CIGAR, Storto and Yang, 2024). CIGAR is a reanalysis system developed by Storto and Yang (2024). Its resolution is 1°×1°. It is based on the NEMO ocean model version 4.0.7 (https://www.nemo-ocean.eu/, last access: November 2025) and is forced by the ERA5 atmospheric reanalysis from the European Center for Medium-range Weather Forecast (ECMWF) (https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5, last access: November 2025). The system uses a three-dimensional variational (3D-Var) scheme to assimilate in situ T/S profiles from the EN4 data set. Notably, CIGAR does not assimilate satellite altimetry data, which allows for independent comparisons. The system consists of 32 ensemble members generated through varying configurations (e.g., perturbed initial conditions and atmospheric forcing). In this study, we use the ensemble mean of the 32 members to derive the thermosteric, halosteric and steric sea level.

Bouih et al. (2025) considered the ensemble mean of five ocean reanalyses over 2004–2019, including CIGAR, to compute the manometric component and showed that this ensemble mean and CIGAR manometric components are highly similar. For that reason, here we only consider CIGAR, but extend it until 2022.

Note that, while the CIGAR reanalysis ingests the same T/S measurements as SIO and EN4 (specifically using the EN4 raw in-situ profile dataset), the reanalysis adopts different quality control, observational sub-sampling, mapping (horizontal correlation) methods, and vertical propagation. Additionally, in the reanalysis, other factors connected to the model physics (e.g., vertical mixing) and atmospheric and river forcing further shape the time evolution of the ocean state. The reanalysis thus provides a generally different realization of the ocean state evolution than the other products. Vertical covariances and dynamical processes contribute to the evolution of the deep ocean, unlike observation-only products.

The thermosteric, halosteric, and total steric sea level time series are computed from the gridded temperature and salinity products (both reanalyses and gridded observation products) using the Gibbs SeaWater (GSW) Oceanographic Toolbox (McDougall and Barker, 2011), which implements the Thermodynamic Equation Of Seawater (TEOS-10) standard. Specifically, 3D density anomalies are vertically integrated to obtain the total steric sea level; thermo- and halo- steric sea levels are obtained similarly, using density anomalies at constant salinity or temperature, respectively (see https://link.springer.com/article/10.1007/s00382-015-2554-9, last access: November 2025).

Manometric sea level change generally refers to the sea-level component associated with ocean mass variations. It is driven by water mass exchange with the continents, such as land ice melt and terrestrial water storage changes, as well as by mass redistribution within the ocean driven by ocean circulation. In this study, because the global mean ocean mass trend is removed, the manometric sea level refers to the regional redistribution of ocean mass around the global mean. In this study, we utilize two methods to obtain an estimate of the manometric component. One method is to use the ensemble mean of the GRACE mascon solutions from CSR, JPL, and GSFC to obtain the GRACE-based manometric estimate. The other method, an alternative to using GRACE solutions, relies on ocean reanalysis data and follows the approach developed by Camargo et al. (2023). Here, we use the sterodynamic sea level provided by the CIGAR ocean reanalysis, which combines steric and dynamic sea level changes. To isolate the manometric component, at every grid mesh, we subtract the local steric value from the sterodynamic variable, both being computed at the same grid mesh. We further add the contemporary Gravitational, Rotational, and Deformation (GRD) fingerprints (representing solid Earth deformations and gravitational effects due to present-day land ice melt and terrestrial water storage changes; Gregory et al., 2019) to obtain a manometric component comparable to GRACE. The sea level fingerprint data used in this study are based on monthly GRD fingerprint grids estimated by Adhikari et al. (2019). Since this original dataset ends in 2016, we extended the time series up to 2022 using linear extrapolation, assuming that the observed trend remains constant after 2016 (see Bouih et al., 2025 for details).

This reanalysis-based estimate is hereafter referred to as CIGAR manometric sea level.

As mentioned above, the CIGAR reanalysis assimilates in-situ T/S profiles. This would mean that the steric and manometric sea levels are not fully independent within the reanalysis due to their dynamical interaction. However, the CIGAR-based manometric component is estimated from the sterodynamic variable that includes both ocean mass redistribution estimated by the model and steric contribution (the latter being further subtracted from the sterodynamic variable). Because the reanalysis is free to adjust the partitioning of sea level variations into steric and manometric components according to the ocean physics and the assimilation framework, without altimetry ingestion, the manometric component constitutes a quasi-independent realization.

To ensure spatial consistency across all observing systems, all gridded data sets were spatially interpolated onto a 1°×1° grid and averaged at monthly interval. Post-processing was performed on such 1°×1° monthly grids. A three-month moving average filter was applied to the corresponding time series. For the spatial analysis, a common mask was applied to all gridded components to exclude regions with high uncertainty. This mask covers latitudes from 66° S to 66° N, excludes inland seas, and omits coastal regions where the distance from land is less than 300 km (see Bouih et al., 2025 for details). For each data set and at each grid mesh, seasonal signals (annual and semi-annual) are removed from the time series through a least-squares adjustment of 6 and 12-month sinusoids, as well as the globally averaged trend of each data set computed over the study period. The study period spans from January 2004 to December 2022. To allow consistent comparison and budget combination across datasets, all gridded anomaly fields and time series were systematically re-referenced to a common baseline period (2004–2009) by subtracting their respective temporal mean over this timeframe.

For the altimetry data, we use the uncertainties estimated by Prandi et al. (2021). For the GRACE-based manometric component, uncertainties can be derived from the dispersion of the data sets with respect to the ensemble mean. While formal error estimates are provided with individual mascon solutions (e.g., Watkins et al., 2015), they primarily quantify inversion noise and lack spatial error covariance information, leading to overly optimistic basin-wide uncertainties (<0.1 mm yr⁻¹). Furthermore, as end-users, we have no a priori ground to consider one mascon solution superior to the others. We therefore adopt the ensemble mean of the JPL, CSR, and GSFC solutions as our main manometric dataset. In accordance with established community standards for sea level budget assessments (Barnoud et al., 2021; Blazquez et al., 2018; WCRP Global Sea Level Budget Group, 2018), we estimate the manometric uncertainty from the ensemble dispersion across the three processing centers, which naturally captures the structural processing uncertainty among these equally plausible products.

For the CIGAR-based manometric, thermosteric, halosteric and steric data, the dispersion of the 32 realizations around the ensemble mean was used to estimate the uncertainties.

For the residual time series, the uncertainty was estimated by quadratically summing the error of each term of the budget equation. Regarding the trend uncertainty, it is expressed as the standard error of the least-squares fit. To obtain the uncertainty at the 95 % confidence level, we scaled the standard error by a factor of 2 (representing the 2-sigma interval).

Figure 1a–d show the thermosteric sea level (TSL) trends over 2004–2022 over the North Atlantic for the three Argo products SIO, JAMSTEC and EN4, and CIGAR (0–2000 m) reanalysis data. Similarly, Fig. 1e–h shows the halosteric sea level (HSL) trends for the same data sets and period. As expected, due to thermohaline compensation particularly strong in the North Atlantic (e.g., Pardaens et al., 2011; Wang et al., 2010; Wunsch et al., 2007), the two components display opposite trends. Regarding the thermosteric component (Fig. 1a–d), all four products display similar spatial patterns, especially indicating significant positive signals in the western North Atlantic. Regarding the halosteric component (Fig. 1e–h), the four products share similar spatial patterns; however, the magnitude of the SIO product is slightly smaller than the other three in the Northeast Atlantic, the others being very similar. This difference likely results from the SIO processing that corrects for the instrumental salinity drift reported in Argo floats since 2016, whereas the other products do not account for this. The steric sea level (SSL) trends are shown in Fig. 1i–l. The magnitude of the total steric change is smaller than that of the thermosteric component alone. This implies that the halosteric contribution plays a significant compensatory role in the North Atlantic (consistent with previous published studies; e.g., Bouih et al., 2025; Llovel and Hochet, 2025). Unlike in the case of the global mean, the influence of salinity changes on sea level cannot be ignored in this region.

Here we use the CIGAR reanalysis data to compute thermosteric, halosteric, and steric sea level changes for the deep ocean (below 2000 m). We consider two cases: (1) SIO upper ocean (0–2000) combined with CIGAR deep ocean (below 2000 m). Hereafter it is called SIO hybrid. (2) CIGAR full depth. Trends of thermosteric, halosteric, and steric full depth for SIO hybrid and CIGAR full depth over 2004–2022 are shown in Fig. 2a, c, e for SIO hybrid, and Fig. 2b, d, f for CIGAR full depth. Both products display strong positive thermosteric and negative halosteric signals in the western North Atlantic. Small difference is seen for the steric component in the regions of the subtropical gyre between 20–40° N, with larger signal near 30° N in the CIGAR full depth case.