Global and regional ocean mass budget closure since 2003

The study addresses discrepancies between GRACE/GRACE-FO satellite gravimetry estimates of ocean mass and steric-corrected satellite altimetry, especially a post-2015/2016 flattening in GRACE-derived ocean mass trends that has led to concerns about systematic biases in sea-level observations. Precise quantification of contributions to global and regional sea level change is critical given societal vulnerability to coastal sea-level rise. Ocean mass change primarily reflects land-to-ocean water transfer from land ice loss and terrestrial water storage changes. The authors aim to provide three largely independent monthly ocean mass datasets for 2003–2022—GRACE/GRACE-FO, steric-corrected altimetry, and a reconstruction (OMrecon) from land sources—to evaluate closure of the ocean mass budget globally and regionally, accounting for recent corrections to wet troposphere correction and halosteric sea level drift.

Multiple studies have attempted to close the sea-level and ocean-mass budget globally and regionally by comparing GRACE with steric-corrected altimetry. Closure was typically achieved over 2005–2015, when GRACE operated nominally and Argo reached full capacity, but regional closure remained partial. More recent analyses did not confirm the GRACE-GFO post-2016 flattening and suggested GRACE-GFO underreported ocean mass change following 2016 instrument issues. Since 2016, closure between steric-corrected altimetry and GRACE-GFO did not recover, prompting questions about observational techniques and global water cycle understanding. The literature also highlights issues including ARGO salinity drift affecting halosteric estimates, sensitivity to glacial isostatic adjustment solutions, and regional dynamic mass discrepancies in reanalysis versus GRACE-derived dynamics, notably in the Atlantic.

The authors develop a monthly 0.5° gridded ocean mass reconstruction (OMrecon) for 2003–2022 by summing five land-to-ocean flux components and applying sea-level fingerprints (GRD effects) using the ISSM-SEESAW framework with ensemble uncertainty propagation:

• Land ice: Greenland Ice Sheet (incl. peripheral glaciers), Antarctic Ice Sheet (incl. peripheral glaciers), and global glaciers (excluding ice-sheet peripheries). Antarctic mass change is based on IMBIE region-averaged estimates combined with GRACE-derived spatial distribution and seasonality; half of Antarctic peripheral glacier mass balance is added, divided across regions. Antarctic mass is extended through 12/2022 using GRACE-FO via an extension model.
• Greenland: GEUS daily basin-scale mass balance is downscaled to 0.5° monthly; peripheral glaciers from the glacier model are added; differences with GRACE are used to estimate peripheral glacier contributions; extension follows the same scheme as glaciers.
• Glaciers: A global glacier model driven by ERA5 is used; due to higher modeled loss (−328 Gt/yr) than observation-based estimates (−230 Gt/yr), model output is scaled by 0.71 to align with observations. Mass balance is extended to 12/2022 using a decomposition plus detrended, non-seasonal GRACE-FO signal per grid cell: vE(t) = β + αt + A·sin(ωt) + gE(t). A filtering kernel separates GRACE-observed glacier signals from overlapping LWS before extension.
• Land Water Storage (LWS): WaterGAP2.2e (ERA5-forced) provides total LWS, with human-induced impacts (reservoirs, water use/groundwater depletion) distinguished via a neutralized run to separate LWS Human and LWS Natural.
• Atmospheric mass: Global mean atmospheric water mass is subtracted from the land-source sum to form OMrecon (barystatic + atmosphere).

GRD-induced sea-level change: For each source, 1000 ensemble members are generated assuming Gaussian measurement uncertainties; GRD responses are computed monthly with and without rotational feedback (seasonal fingerprints exclude rotational feedback to match processed GRACE/altimetry). Mass conservation is enforced with a correction factor (361.8 Gt ≈ 1 mm global sea level). Outputs include vertical deformation (for OBD), relative and absolute sea level, gridded at 0.5°.

GRACE observations: GSFC mascons RL06 with GAD atmospheric/non-tidal ocean removal over oceans; GIA corrected with ICE6G-D. Time series are interpolated to mid-month; missing months are filled after seasonal removal (seasonality restored). The 13-month GRACE–GFO gap is retained.

Steric-corrected altimetry: Along-track SLAs from Jason-1/2/3 and Sentinel-6 are corrected with a Wet Troposphere Correction (WTC): default MWR is used pre-2016 and merged with a CDR-based WTC during Jason-3 (2016–2022) to address stability. Intermission biases are corrected via tandem periods. SLAs are gridded monthly at 0.5°. Uncertainty combines grid-cell variability and intermission bias. GIA effects (absolute sea level) and Ocean Bottom Deformation from loading are removed from altimetry to enable mass comparisons. Steric sea level (thermosteric + halosteric) is computed from gridded T/S analyses (0–5400 m) using HOMaGE/TEOS-10; to correct ARGO salinity drift since 2015, the annual global mean halosteric low-pass signal is removed while preserving sub-annual variability. The deep (>2000 m) steric contribution is assessed; limiting to <2000 m reduces global steric trend by 0.10 ± 0.04 mm/yr.

Masks and dynamics: A narrow 50 km coastal cutoff is applied to retain most dynamic sea-level signals present in GRACE and altimetry; very shallow areas (<200 m depth) and regions around the 2004 Sumatra and 2011 Sendai earthquakes are masked.

Trend/seasonality estimation: Trends, annual phase, and amplitude are derived via 10,000-sample bootstrapping of inverse-variance-weighted time series, reporting means and 1σ uncertainties.

• Over 2003–2022, OMrecon indicates a total global ocean mass increase of 44.6 ± 3.8 mm water level equivalent, largely attributable to land ice loss and human water management. Contributions: Greenland 17.0 ± 0.8 mm, Antarctica 8.6 ± 1.4 mm, Glaciers 13.4 ± 1.4 mm, LWS Human 7.0 ± 1.0 mm, LWS Natural 0.2 ± 2.8 mm (negligible over 20 years). Approximately 85% of the reconstructed mass increase is due to ice loss.
• Linear trend estimates (2003–2022): OMrecon (barystatic + atmosphere) 2.23 ± 0.19 mm/yr; GRACE 2.11 ± 0.14 mm/yr; steric-corrected altimetry with MWR/CDR WTC 2.28 ± 0.36 mm/yr; with MWR-only WTC 2.39 ± 0.34 mm/yr. Barystatic sum (ice + LWS) 2.32 ± 0.18 mm/yr; global mean atmospheric mass −0.09 ± 0.02 mm/yr.
• Cumulative changes (2003–2022): GRACE 42.2 ± 2.8 mm; steric-corrected altimetry 55.8 ± 7.2 mm without halosteric drift correction, 47.8 ± 7.2 mm with halosteric correction; adopting CDR-based WTC during Jason-3 yields 45.7 ± 7.3 mm. With these corrections, the three estimates agree within 1σ.
• Seasonal cycle: Dominated by natural LWS; OMrecon seasonal phase agrees within ~10° of GRACE and altimetry, with OMrecon showing a larger amplitude. Ice and human LWS contribute a minor (~10%) share to seasonality.
• Interannual variability: OMrecon underestimates interannual variability relative to GRACE and altimetry, particularly during ENSO extremes; maximum 12-month averaged rates up to ~10 mm/yr and minima ~−5 mm/yr observed. Post-2019 slowdown aligns with persistent La Niña and a temporary Antarctic Ice Sheet mass gain driven by enhanced surface mass balance (2020–2022).
• 2020–2022 deviation: OMrecon trends are lower than GRACE/altimetry by 1.75 ± 0.18 mm/yr, attributed to hydrological model underestimation of LWS increases during La Niña, notably in mid-latitude Africa (−1500 Gt bias; ~−4 mm WLE over 2019–2022).
• Regional budgets: All major basins show significant ocean mass gains with trends ~1.40–2.85 mm/yr. Differences between GRACE and steric-corrected altimetry vary regionally, especially between North and South Atlantic, likely due to regional halosteric drift corrections, dynamic mass differences, and GIA model sensitivities. Combining North and South Atlantic restores basin-scale closure within uncertainties.
• Equal-weighted average of the three datasets gives a global ocean mass trend of 2.21 ± 0.25 mm/yr (2003–2022).

The study demonstrates that, after correcting for ARGO halosteric salinity drift and using a more stable CDR-based wet troposphere correction for altimetry during Jason-3, the global and regional ocean mass budget closes over 2003–2022 among three largely independent estimates (GRACE/GRACE-FO, steric-corrected altimetry, and OMrecon). The findings support the accuracy of GRACE-FO in monitoring ocean mass changes; prior non-closure since 2016 is more plausibly due to MWR calibration issues in altimetry and sensitivities to GIA corrections, compounded by ARGO salinity drift. Over long-term and seasonal timescales, OMrecon aligns well with GRACE-FO, while underestimating interannual variability, particularly during ENSO phases, due to hydrological model limitations in capturing rapid LWS fluctuations. Regionally, discrepancies (e.g., in the Atlantic and Indian Ocean) underscore the influence of regional halosteric biases, dynamic mass redistribution, and GIA model choices. Overall, the results attribute nearly all long-term ocean mass increase to land ice wastage and human water management, with natural LWS dominating variability but showing negligible long-term trend.

By reconstructing ocean mass from land-based sources and comparing it with GRACE/GRACE-FO and steric-corrected altimetry, the study achieves closure of the global and regional ocean mass budget over 2003–2022 when applying updated halosteric drift corrections and CDR-based WTC. The long-term rise in ocean mass is driven primarily by land ice loss and secondarily by human-induced land water storage changes; natural LWS drives seasonal and interannual variability without contributing a significant long-term trend. The recent apparent slowdown since 2020 is attributed to a prolonged La Niña and a temporary increase in Antarctic surface mass balance. Future work should focus on improving calibration and stability of altimetric WTC, refining ARGO salinity drift corrections and regional halosteric adjustments, reducing GIA-related uncertainties, and enhancing hydrological models’ responsiveness to short-term climate variability to better capture interannual LWS changes.

• OMrecon includes extrapolated mass balance estimates for Antarctica (24 months) and glaciers (up to 54 months), adding uncertainty despite methodological safeguards.
• Hydrological model (WaterGAP2.2e) driven by ERA5 underestimates precipitation in some mid-latitude regions (e.g., Africa), leading to underestimation of LWS during recent La Niña and underrepresentation of interannual variability.
• Deep-ocean steric estimates are limited by sparse observations below 2000 m; regional deep steric contributions are uncertain.
• GIA model choice significantly impacts GRACE-derived regional trends, especially in the North Atlantic; related uncertainties may not be fully captured in GRACE error budgets.
• OMrecon does not include dynamic ocean mass redistribution (zero global mean but regionally significant), while GRACE and altimetry do; coastal and very shallow regions are masked, potentially omitting relevant signals.
• GRACE data gaps (e.g., 13-month hiatus) and interpolation of missing months introduce additional uncertainty.
• Glacier model scaling and separation of glacier/LWS signals via filtering kernel introduce methodological uncertainty.