Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 65 of the 65 references it cites were found in a registry or an index; none of the 23 sources its claims cite is checked for support yet.

SWOT Discharge Accuracy Benchmarked in South America

Open access

Abstract

The Surface Water and Ocean Topography (SWOT) satellite mission offers unprecedented estimates of global river discharge. This study provides a first continental‐scale assessment of SWOT Consensus discharge performance across South America and compares it with a large‐scale hydrodynamic model and in situ observations. Consensus generally exhibits higher skill with a median Kling‐Gupta Efficiency (0.43) across the 569 benchmarked reaches. Performance varies systematically with physiographic settings. Random Forest and Spearman correlation analyses identify slope and distance to outlet as the top‐ranked controls among the variables tested, though no single variable explains performance, reflecting its multi‐dimensional nature. Consensus discharge inherits systematic bias from its machine learning prior, and substantial unexplained variability highlights that further work is needed to fully diagnose low‐skill reaches.

Overview

CoChat Summary

TL;DR

This study provides a first continental-scale benchmark of SWOT Consensus river discharge across South America against in situ gauges and the HyMAP hydrodynamic model over 569 reaches. Consensus shows moderate skill (median KGE 0.43) and generally outperforms HyMAP (median KGE 0.22), but performance is spatially heterogeneous and inherits bias from its machine-learning prior.

Key findings

  • Across 569 benchmarked reaches, SWOT Consensus had median KGE = 0.43, with 40.25% of reaches at KGE ≥ 0.5, median r = 0.85, median NSE = 0.37 (63% of reaches NSE ≥ 0), and median |nBIAS| = 27% (IQR 12%–60%). HyMAP had median KGE = 0.22, 22% of reaches at KGE ≥ 0.5, median r = 0.75, median NSE = −0.12, and median |nBIAS| = 39% (IQR 17.8%–67.1%).
  • Failures were only partially co-located: 46 of 569 reaches (8.1%) failed in both products (KGE < −0.41), concentrated in Brazil, Chile, and Argentina. Where Consensus failed, HyMAP performed adequately at 55 of 101 reaches (54.5%); where HyMAP failed, Consensus performed adequately at 67 of 113 reaches (59.3%).
  • Slope and distance to outlet were the top-ranked controls among nine tested variables. Spearman correlations with KGE were slope ρ = −0.37, distance to outlet ρ = 0.34, width ρ = 0.32, and drainage area ρ = 0.31; Random Forest cross-validated R² = 0.30, ρ = 0.47, with top mean absolute SHAP values for slope (0.175) and distance to outlet (0.158). No single variable explained performance.
  • Consensus bias tracked the GRADES-hydroDL prior: across 530 shared reaches, mean simulated discharge correspondence was ρ = 0.90 (p < 0.001), bias direction was shared in 74% of reaches, and normalized bias was correlated at ρ = 0.61 (p < 0.001). At low flows HyMAP had better relative errors; at higher flows Consensus had lower absolute and relative errors, with the bulk of observations in the 100–1,000 m³/s bin where Consensus performed best.

Methods

  • SWOT Consensus discharge came from the Confluence workflow using version 3 of the SWOT Level 4 Sword of Science River Discharge Product (unconstrained), covering 31 March 2023 to 2 May 2025, with SWORD v16, SWOT Level 2 River Single-Pass Vector Data Product version C, and δHBV2 and GRADES-hydroDL as machine-learning priors. Consensus was the median of algorithm-derived discharge estimates after excluding those with coefficient of variation ≥ 0.5; four FLPE algorithms were used because HiVDI and SADS were not operational.
  • HyMAP simulated discharge across South America at 0.05° (~5 km) for 2023–2025, forced by GLDAS-2.2 runoff, with physically based, uncalibrated parameters and MERIT-Hydro river network/elevations.
  • In situ daily discharge came from 1,909 gauges across eight countries: Brazil 723, Chile 449, Colombia 337, Argentina 208, Peru 132, Uruguay 34, Paraguay 4 from EBY plus 22 from HyBAm. Gauges were matched to SWORD v16 reaches, retaining 569 pairs with ≥ 10 temporally coincident observations across all products; SWOT averaged 34.1 estimates per reach over 763 days (about one every 22 days, range 5–76 days). Analysis used KGE, NSE, bias, RMSE, Spearman correlation, and Random Forest with 5-fold cross-validation and SHAP on nine predictors.

Limitations

  • Stated: Consensus discharge was available for only 44% of South American reaches; the Amazon mainstem was excluded from the benchmark and analyzed separately. The complete dataset represents the best SWOT observations after quality filters, averaging 4.46% of observed days over the study period, while HyMAP provides daily estimates. Discharge data were lacking for many large rivers (Amazon estimates for 50% of reaches, La Plata 39%), and gauge records carry their own uncertainties. The nine variables did not fully explain performance; unexplained variability likely reflects local hydraulic complexity, hydrologic regulation, geomorphological settings, or internal prior bias. Results are relative tendencies across the benchmarkable subset, not absolute rankings.
  • Stated: HyMAP was forced by GLDAS-2.2 runoff, and input runoff fields largely control simulated discharge accuracy, so a different runoff dataset could yield substantially different results; HyMAP parameters were not calibrated. Consensus inherits systematic bias from its machine-learning prior.
  • Evident from the text: temporal sampling is sparse (about one SWOT observation every 22 days on average, and only 4.46% of observed days after filtering), limiting assessment of daily dynamics; only four FLPE algorithms were operational, so Consensus excludes HiVDI and SADS; KGE values below −2 were capped at −2, and basin correlations capped KGE below −10 at −10, which may affect tail characterization.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1029/2026gl122871

Cites

65 of 65 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1
Figure 1 CC-BY-NC · From the PDF
Image 2
CC-BY-NC · From the PDF
Figure 2 Spatial distribution of KGE performance across 569 gauges in SA (a) Consensus (IQR = 0.65), (b) HyMAP (IQR = 0.76), and (c) their difference (ΔKGE). Panel (d) shows agreement between Consensus and HyMAP KGE performance within HydroSHEDS Level 3 basins. For each basin, Pearson correlation was calculated across stations after setting KGE values below 10 to 10 to reduce the influence of extreme negative tail values. The map shows the resulting within-basin correlation between Consensus and HyMAP KGE values, reflecting agreement in the relative pattern of gauge performance. The scatterplots to the right show paired gauge-level KGE values relative to the 1:1 line and the median KGE for each product, providing complementary information on absolute agreement in skill. For readability, the scatterplot axes are truncated at KGE = 2; full basin scatterplots are provided in Figure S7 of Supporting Information S1.
Figure 2 Spatial distribution of KGE performance across 569 gauges in SA (a) Consensus (IQR = 0.65), (b) HyMAP (IQR = 0.76), and (c) their difference (ΔKGE). Panel (d) shows agreement between Consensus and HyMAP KGE performance within HydroSHEDS Level 3 basins. For each basin, Pearson correlation was calculated across stations after setting KGE values below 10 to 10 to reduce the influence of extreme negative tail values. The map shows the resulting within-basin correlation between Consensus and HyMAP KGE values, reflecting agreement in the relative pattern of gauge performance. The scatterplots to the right show paired gauge-level KGE values relative to the 1:1 line and the median KGE for each product, providing complementary information on absolute agreement in skill. For readability, the scatterplot axes are truncated at KGE = 2; full basin scatterplots are provided in Figure S7 of Supporting Information S1.CC-BY-NC · From the PDF
Image 4
CC-BY-NC · From the PDF
Figure 3 Results of the correlation and Random Forest analysis (a) Spearman rank correlation matrix between metrics (KGE, its components r, |α 1| and |β 1|, and NSE) and nine predictor variables. (b) SHAP summary plot showing predictor effects on random forest KGE predictions. Higher SWOT slope, higher relative width uncertainty, and higher badness score generally shifted predictions toward lower KGE, while distance to outlet and drainage area also contributed to model behavior, reflecting differences in basin setting across the continent. Spearman correlation graphs can be found in Figure S8 and S9 of Supporting Information S1.
Figure 3 Results of the correlation and Random Forest analysis (a) Spearman rank correlation matrix between metrics (KGE, its components r, |α 1| and |β 1|, and NSE) and nine predictor variables. (b) SHAP summary plot showing predictor effects on random forest KGE predictions. Higher SWOT slope, higher relative width uncertainty, and higher badness score generally shifted predictions toward lower KGE, while distance to outlet and drainage area also contributed to model behavior, reflecting differences in basin setting across the continent. Spearman correlation graphs can be found in Figure S8 and S9 of Supporting Information S1.CC-BY-NC · From the PDF
Image 6
CC-BY-NC · From the PDF

Verification

Claims the paper makes

0 Valid · 0 Not valid · 14 Not checked · 62 Own results and interpretations

18 more claims

Create a free account to see all 76 claims and to verify the 18 references still open.

See all claims — free account
  • It may state two separate claims as one. 45
  • Probably not a substantial claim. 14
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Citations

Cites

Show all 65 references

Research works printed but not linked: 4

  1. Getirana A., 2025, The HyMAP‐3 Hydrodynamic Model (Technical Memorandum NASA/TM‐20250009546), 30
  2. JPL, 2018, SWOT science requirements document (Rev. B) (science requirements document JPL D‐61923)
  3. JPL, 2025, SWOT Science Data Products User Handbook (D‐109532, Revision A)
  4. UN Water, 2024, Water for prosperity and peace

Cited by

See all 2

Metadata

Authors
Jover, L., Getirana, A., Gleason, C., An, H., Crisanti, S., Dose, E., Eylander, J. B., Hellwig, I., Lavado, W., Martinez, J., Lozano, J. L. S., Santini, W., Sosa, M., & Lakshmi, V.
Affiliations
  • Johns Hopkins University (US)
  • Goddard Space Flight Center (US)
  • University of Massachusetts Amherst (US)
  • United States Army (US)
  • Yacyretá Binational Entity Asunción Paraguay
  • United States Army Corps of Engineers (US)
  • Subsecretary of Water Resources Buenos Aires Argentina
  • National Meteorological and Hydrological Service Lima Peru
  • Centre National de la Recherche Scientifique (FR)
  • Water Resources Engineer Aquaveo LLC Provo UT USA
Corresponding author
Jover, Laila
Venue
Geophysical Research Letters
Published
2026-09-08
Volume
53
Issue
17
ISSN
  • 0094-8276 (ISSN-L)
  • 1944-8007
OpenAlex
W7211943190
Cited by
2
References
65
Citations by year
2026: 2
FWCI
2.41
Type
journal article
Language
en
Open access
Licence
Locations
  • Geophysical Research Letters (publishedVersion, cc-by-nc)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
U.S. Army Corps of Engineers (W912HZ25C0013)
Topics
  • Physical Sciences › Environmental Science › Water Science and Technology › Hydrology and Watershed Management Studies
  • Physical Sciences › Environmental Science › Environmental Engineering › Soil Moisture and Remote Sensing
  • Physical Sciences › Environmental Science › Global and Planetary Change › Flood Risk Assessment and Management
Keywords
SWOT mission, river discharge estimation, South America, Kling-Gupta efficiency, in situ observation, machine learning bias, river slope, random forest

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: SWOT Discharge Accuracy Benchmarked in South America. https://app.cochat.ai/papers/10.1029/2026gl122871
The paper
Jover, L., Getirana, A., Gleason, C., An, H., Crisanti, S., Dose, E., Eylander, J. B., Hellwig, I., Lavado, W., Martinez, J., Lozano, J. L. S., Santini, W., Sosa, M., & Lakshmi, V. (2026). SWOT Discharge Accuracy Benchmarked in South America. Geophysical Research Letters. https://doi.org/10.1029/2026gl122871

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.