Satellite Gravity Tracks River Freshwater Flowing Into the Atlantic Ocean

Freshwater flowing from the world’s great rivers into the Atlantic Ocean has long been one of the weakest links in ocean modeling. Now a research team led by Louis Kern of Magellium in Toulouse, France, has shown that satellite measurements of Earth’s shifting gravity field can be combined with atmospheric data to estimate how much water 18 major rivers deliver to the Atlantic, and that using these estimates in an ocean model measurably improves the simulation of sea surface salinity, upper-ocean circulation and meridional transports. The study, published in the journal Ocean Science, evaluates the impact of riverine freshwater fluxes on the Atlantic Ocean by solving the water mass balance equation at the scale of each river basin.

The Water Mass Balance, or WMB, approach rests on a simple accounting identity: the change in land water storage over time equals precipitation minus evapotranspiration minus river discharge. The storage term is the hard part, and it is where the satellites come in. Measurements from the GRACE and GRACE-FO missions, which track how gravity varies as water is redistributed across the planet, allow researchers to estimate terrestrial water storage anomalies from April 2002 to August 2023 at monthly timescales and with a spatial resolution of about 1 degree. The team built an ensemble of 120 GRACE and GRACE-FO solutions drawn from five processing centers, combining different geocenter models, oblateness corrections, glacial isostatic adjustment models and filtering strategies to capture the uncertainties inherent in the gravity data.

Atmospheric fluxes were estimated in two ways. The first, called WMB1, uses precipitation and evapotranspiration from the ERA5 atmospheric reanalysis. The second, WMB2, pairs GPCC observational precipitation with evapotranspiration from the ISBA-CTRIP hydrological model, which is itself forced with GPCC precipitation, ensuring consistency between the atmospheric fluxes. The difference between net precipitation and the time derivative of storage change then yields river discharge for each basin. The researchers selected rivers with average discharge exceeding 2,000 cubic meters per second and catchment areas larger than 100,000 square kilometers, the latter chosen to match the roughly 300 by 300 kilometer resolution of the gravity measurements. The 18 resulting basins span North and South America, Africa and Europe, and together carry a long-term average discharge of 389,411 cubic meters per second, about 64 percent of the total discharge to the Atlantic recorded in the HydroRIVERS database.

Validation against in situ river gauges produced strikingly mixed results. For the Amazon, the largest freshwater contributor to the Atlantic, the method performed remarkably well. Comparing WMB estimates with measurements at the Óbidos gauging station, roughly 500 kilometers upstream of the delta, the Pearson correlation coefficient reached 0.87 with ERA5 forcing and 0.88 with GPCC and ISBA-CTRIP. The mean discharge of 1.734 times ten to the fifth cubic meters per second was captured within 2 percent by WMB1 and within 3 percent by WMB2, and the annual and monthly climatology could be estimated with an error of less than 5 percent. The WMB1 configuration achieved a modified Kling-Gupta Efficiency score of 0.84, indicating excellent agreement, and correctly reproduced May and November as the months of highest and lowest discharge.

Yet the same method faltered elsewhere. The WMB2 configuration overestimated the Amazon’s annual discharge cycle by 46 percent, a discrepancy traced to lower mean values, higher noise and greater variability in ISBA evapotranspiration compared with ERA5. Across the full set of rivers, the choice of atmospheric dataset produced enormous spreads: the Congo River’s mean discharge differed by 53 percent between the two configurations, the Niger by 74 percent, and the Magdalena by 87 percent. In the extreme case of the São Francisco River, WMB2 produced a negative mean discharge, a clear signal that the method failed to capture the river at all. The authors attribute these failures to excessive noise or biases in one or more components of the water balance equation, emphasizing the method’s sensitivity to the quality of regional atmospheric data.

Performance also varied sharply by continent. The WMB approach yielded reasonable results for South American rivers, particularly the Amazon, when forced with ERA5, with proportional bias values close to zero and correlation scores above 0.5. Results for African and North American rivers were more variable and generally less satisfactory. Both configurations struggled with African rivers: WMB1 typically underestimated their mean discharge by 31 to 112 percent, while WMB2 overestimated by 45 to 580 percent. Correlation was higher with ERA5 than with GPCC-ISBA for 8 of the 11 rivers analyzed. A persistent weakness across all rivers was exaggerated high-frequency variability: the median variability ratio, comparing the spread of WMB estimates to in situ data, was 2.2 for WMB1 and 1.9 for WMB2, meaning the satellite-derived discharges fluctuated far more than the gauges recorded.

The method does offer one decisive advantage over gauges alone: spatial coverage. In situ measurements were available for only 11 of the 18 rivers since 2000, and gauging stations are usually located far upstream of river mouths. The Óbidos station’s drainage area is about 22 percent smaller than the full Amazon basin, excluding major tributaries such as the Tapajos, Xingu and Jari, so the mean discharge measured there is 15 percent lower than at the river mouth, and the annual cycle amplitude is underestimated by 20 percent. Relying solely on the 11 gauging stations would underestimate the total mean freshwater flux by 21 percent and the seasonal amplitude by 27 percent relative to estimates at the outlets. For context, a 10 percent underestimation of the Amazon’s mean discharge is equivalent to omitting the entire flow of the Mississippi River.

The real test came when the WMB discharge estimates were substituted for the standard climatological river discharge dataset in simulations using the NEMO ocean model, version 4.0.7, configured on a 1-degree ORCA1 grid with 75 vertical levels and forced by ERA5 meteorology. The experiments, covering January 2003 to December 2020 and initialized from the same CIGAR reanalysis conditions, included no data assimilation, allowing the impact of the river discharge dataset to be evaluated without interference from observations. When climatological estimates were replaced by the WMB approach, sea surface salinity biases decreased at the Amazon outlet and across the whole Atlantic. The Atlantic basin mean reduction in the mean absolute error of sea surface salinity, evaluated against the EN4 observational analysis, was 0.07 psu for the ERA5-driven experiment and 0.08 psu for the GPCC-ISBA version. Using WMB estimates for only the Amazon, Orinoco and Tocantins rivers, the three with high-confidence results, achieved about half of that bias reduction.

The improvements extended below the surface. Salinity profiles within 500 kilometers of the shoreline showed that the WMB experiments significantly reduced the fresh bias in the surface layer, with improvements reaching roughly 100 meters deep, into the halocline where riverine freshwater influence is most pronounced. The overall reduction in root mean square error in the top 100 meters was 10 percent for the ERA5 configuration and 11 percent for GPCC-ISBA, with the three-river experiments and a daily-varying runoff alternative each achieving 5 percent. An error decomposition showed that a significant portion of the RMSE reduction was associated with bias correction rather than random error reduction, suggesting the climatological runoff had introduced systematic deviations that the WMB dataset effectively mitigated. The reduced freshwater input weakened near-surface haline stratification in the plume regions, allowing vertical mixing to redistribute momentum over a deeper layer, which reduced surface current speeds downstream of the Amazon and Orinoco mouths while intensifying flow at the river mouths and in the equatorial band. Meridional transports increased in the South and Equatorial Atlantic up to about 15 degrees north and decreased slightly farther north, though no significant impact was found on the thermohaline overturning circulation over the relatively short simulation period.

The authors conclude that the Water Mass Balance approach, particularly when constrained with ERA5 atmospheric data, is a promising tool for improving ocean model forcing in regions lacking dense hydrological monitoring, but that its accuracy depends heavily on the quality of atmospheric flux inputs and remains limited by coarse spatial resolution and noise, especially in smaller or poorly monitored basins. Notably, the performance ranking flipped between scales: ERA5 forcing generally performed better at the river-basin scale, while GPCC-ISBA fluxes yielded slightly lower salinity errors at the ocean-basin scale, a difference the authors attribute to regional biases partially compensating when integrated over an entire ocean basin. Significant improvements are anticipated from future satellite gravity missions, the Next Generation Gravity Mission and the Mass change And Geosciences International Constellation, which aim to deliver higher spatial and temporal resolution and reduced uncertainties in gravity field observations.

Subject of Research: Estimating river freshwater discharge into the Atlantic Ocean using satellite gravimetry and the water mass balance approach, and assessing the impact on ocean model salinity and circulation simulations.

Article Title: Quantifying freshwater fluxes from Major rivers and assessing their impact on the Atlantic Ocean

Article References: Kern, L., Vaujour, T., Pfeffer, J., Storto, A., Szczypta, C., Garric, G., Sirere, C., Larnicol, G., Yang, C., Bourdalle-Badie, R., & Guinehut, S. (2026). Quantifying freshwater fluxes from Major rivers and assessing their impact on the Atlantic Ocean. Ocean Science, 22(5), 3457-3477. https://doi.org/10.5194/os-22-3457-2026

Image Credits: AI Generated

DOI: 10.5194/os-22-3457-2026

Keywords: satellite gravimetry, GRACE, river discharge, Atlantic Ocean, sea surface salinity, water mass balance, ocean modeling, Amazon River, freshwater fluxes, NEMO model, evapotranspiration, ocean circulation