CoChat

Global Reconstruction of Naturalized River Flows at 2.94 Million Reaches

Open access

The DOI matches its Crossref record; 80 of the 80 references it cites were found in a registry or an index; 2 of the 57 sources its claims cite were checked for support.

Abstract

Spatiotemporally continuous global river discharge estimates across the full spectrum of stream orders are vital to a range of hydrologic applications, yet they remain poorly constrained. Here we present a carefully designed modeling effort (Variable Infiltration Capacity land surface model and Routing Application for Parallel computatIon of Discharge river routing model) to estimate global river discharge at very high resolutions. The precipitation forcing is from a recently published 0.1° global product that optimally merged gauge‐, reanalysis‐, and satellite‐based data. To constrain runoff simulations, we use a set of machine learning‐derived, global runoff characteristics maps (i.e., runoff at various exceedance probability percentiles) for grid‐by‐grid model calibration and bias correction. To support spaceborne discharge studies, the river flowlines are defined at their true geometry and location as much as possible—approximately 2.94 million vector flowlines (median length 6.8 km) and unit catchments are derived from a high‐accuracy global digital elevation model at 3‐arcsec resolution (~90 m), which serves as the underlying hydrography for river routing. Our 35‐year daily and monthly model simulations are evaluated against over 14,000 gauges globally. Among them, 35% (64%) have a percentage bias within ±20% (±50%), and 29% (62%) have a monthly Kling‐Gupta Efficiency ≥0.6 (0.2), showing data robustness at the scale the model is assessed. This reconstructed discharge record can be used as a priori information for the Surface Water and Ocean Topography satellite mission's discharge product, thus named “Global Reach‐level A priori Discharge Estimates for Surface Water and Ocean Topography”. It can also be used in other hydrologic applications requiring spatially explicit estimates of global river flows.

Overview

CoChat Summary

TL;DR

This paper presents GRADES, a 35-year (1979–2014) daily and monthly reconstruction of naturalized river discharge at ~2.94 million reaches, using VIC and RAPID with merged precipitation, machine-learning runoff-characteristic calibration/bias correction, and MERIT Basins hydrography. Evaluated against >14,000 gauges, the bias-corrected simulation achieves PBIAS within 20% for 35% of gauges and monthly KGE >0.6 for 29%, with better skill for SWOT-observable reaches.

Key findings

  • The reconstructed record covers ~2.94 million vector flowlines (median length 6.8 km; mean 9.2 km) and unit catchments (median 36.8 km²; mean 45.6 km²), with 35-year daily/monthly simulations from 1979 to 2014.
  • Against >14,000 gauges, after bias correction 35% (64%) have a percentage bias within 20% (50%), and 29% (62%) have a monthly Kling-Gupta Efficiency >0.6 (0.2); bias correction increased the number of gauges with PBIAS within 20% by 11% and shifted the peak monthly KGE from 0.4–0.5 to 0.6–0.7.
  • For SWOT-observable reaches (width >50 m; 211,968 reaches, 8.8% of modeled reaches), evaluated against ~4,000 gauges, 44% (76%) have PBIAS within 20% (50%) and 42% (75%) have monthly KGE >0.6 (0.2).
  • Performance is better in humid than dry regions; dry regions show median PBIAS >100%, median CC <0.5, and median RV >1.5, while humid regions show median CC >0.6, median PBIAS <20%, and median KGE >0.5. Dams are more influential than natural lakes, though after bias correction differences between gauges with and without lake/dam influence become less prominent.

Methods

  • VIC land surface model at 0.25° (~0.24 million grid cells) was forced by a 0.1° 3-hourly merged gauge/reanalysis/satellite precipitation product and CFSR other variables; it was calibrated grid-by-grid with SCE-UA against machine-learning-derived runoff characteristics (baseflow index, QMEAN, Q10, Q90), then bias-corrected with a new “sparse CDF matching” approach against nine Qc maps.
  • RAPID Muskingum routing was applied to ~2.94 million vector flowlines/unit catchments derived from MERIT Hydro at 3-arcsec (~90 m) resolution (60S–90N); k was estimated from power-law width/depth, Manning’s equation (roughness 0.035), and channel length, with x set to 0.3; runoff was mapped area-weighted, with 5-year spin-up and 3-hourly internal time steps.
  • Evaluation used >14,000 gauges with >3 years of data from a compiled database of >17,000 gauges; gauges were snapped to flowlines (>500 m eliminated) and classified as naturalized (11,313), reservoir-influenced (2,073), or lake-influenced (6,316); skill metrics were KGE, CC, PBIAS, and RV at daily and monthly scales, and SWOT-observable reaches were identified using GRWL widths >50 m.

Limitations

  • The paper states future refinements are needed to improve dry-region simulations, add reservoir regulation modules, update the routing method, and add lake process modules; many headwater streams important for biogeochemical cycling may still not be explicitly modeled; the Qc maps used for calibration and bias correction contain observational information from a small subset of the final evaluation gauges and are assumed as reference despite not being free from errors; local errors may remain.
  • The paper states the model does not represent transmission loss in dry regions; water management activities such as diversions and irrigation are neither parameterized nor implicitly incorporated in the Qc maps (only naturalized flows were used for ML training); GRanD accounts only for large dams, not small reservoirs; and the analyses did not account for distance between dams/lakes and gauges.
  • The text makes evident that observational data quality is unknown or inconsistent across sources, the simple Muskingum routing cannot fully resolve daily flow variations, and precipitation forcing uncertainty plus Qc map errors can impede skill improvement and may increase bias at certain gauges.

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

Integrity signals

Registry

Crossref · 10.1029/2019wr025287

Cites

80 of 80 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 The hydrologic modeling workflow and data sets to constrain the global simulation. VIC = Variable Infiltration Capacity; LSM = land surface model; RAPID = Routing Application for Parallel computatIon of Discharge; MSWEP = Multi‐Source Weighted‐Ensemble Precipitation.
Figure 1 The hydrologic modeling workflow and data sets to constrain the global simulation. VIC = Variable Infiltration Capacity; LSM = land surface model; RAPID = Routing Application for Parallel computatIon of Discharge; MSWEP = Multi‐Source Weighted‐Ensemble Precipitation.CC BY · PubMed Central
Figure 2 Schematics of the newly developed bias correction approach. Stars mark the grid cell under bias correction: dynamic LSM runoff time series are used for cumulative density function construction (blue line) and static runoff characteristics maps are used as the reference (red points). LSM = land surface model.
Figure 2 Schematics of the newly developed bias correction approach. Stars mark the grid cell under bias correction: dynamic LSM runoff time series are used for cumulative density function construction (blue line) and static runoff characteristics maps are used as the reference (red points). LSM = land surface model.CC BY · PubMed Central
Figure 3 Global stream gauges used in model evaluation from (a) different gauge networks. Gauges are classified based on if they are influenced by (b) dams and (c) lakes or not. If influenced by dams/lakes, colors show the number of dams/lakes upstream. USGS = U.S. Geological Survey; GRDC = Global Runoff Data Centre.
Figure 3 Global stream gauges used in model evaluation from (a) different gauge networks. Gauges are classified based on if they are influenced by (b) dams and (c) lakes or not. If influenced by dams/lakes, colors show the number of dams/lakes upstream. USGS = U.S. Geological Survey; GRDC = Global Runoff Data Centre.CC BY · PubMed Central
Figure 4 Histogram of (a) stream order and (b) drainage basin area (km2) of the gauges shown in Figure 3a.
Figure 4 Histogram of (a) stream order and (b) drainage basin area (km2) of the gauges shown in Figure 3a.CC BY · PubMed Central
Figure 5 Global streamflow modeling evaluated against over 14,000 gauges (left column: daily; right column: monthly). Results shown are correlation coefficient (a, b), percentage bias (c, d), relative variability (e, f), and the Kling‐Gupta Efficiency (g, h). Note the CC plot has a range of 0 to 1 because only ~1% of gauges have negative CC; the PBIAS plot has asymmetric ranges because of its definition that can result into asymmetric values for dry and wet biases. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias.
Figure 5 Global streamflow modeling evaluated against over 14,000 gauges (left column: daily; right column: monthly). Results shown are correlation coefficient (a, b), percentage bias (c, d), relative variability (e, f), and the Kling‐Gupta Efficiency (g, h). Note the CC plot has a range of 0 to 1 because only ~1% of gauges have negative CC; the PBIAS plot has asymmetric ranges because of its definition that can result into asymmetric values for dry and wet biases. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias.CC BY · PubMed Central
Figure 6 Effectiveness of the bias correction approach against nine runoff characteristics. The spatial plot shows the percentage bias (%) of the original model simulation; (a)–(g) show the CDF and time series change before (blue) and after bias correction (orange) at seven example grid cells. For clarity, only the runoff time series for Year 1979 are shown. CDF = cumulative density function.
Figure 6 Effectiveness of the bias correction approach against nine runoff characteristics. The spatial plot shows the percentage bias (%) of the original model simulation; (a)–(g) show the CDF and time series change before (blue) and after bias correction (orange) at seven example grid cells. For clarity, only the runoff time series for Year 1979 are shown. CDF = cumulative density function.CC BY · PubMed Central
Figure 7 The same as Figure 5, but for the simulation performance after bias correction.
Figure 7 The same as Figure 5, but for the simulation performance after bias correction.CC BY · PubMed Central
Figure 8 Histogram of the (a) CC, (b) PBIAS, (c) RV, and (d) KGE skill changes before (blue) and after BC (red bars) for the monthly simulation. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias; BC = bias correction.
Figure 8 Histogram of the (a) CC, (b) PBIAS, (c) RV, and (d) KGE skill changes before (blue) and after BC (red bars) for the monthly simulation. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias; BC = bias correction.CC BY · PubMed Central
Figure 9 Boxplot of the model performance statistics based on (a) dam regulation and (b) lake impacts. The left (right) column in each plot shows gauges with (without) the influence from dams or lakes. Statistics for experiment with no BC and that with BC are shown in green and blue boxes, respectively. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias; BC = bias correction.
Figure 9 Boxplot of the model performance statistics based on (a) dam regulation and (b) lake impacts. The left (right) column in each plot shows gauges with (without) the influence from dams or lakes. Statistics for experiment with no BC and that with BC are shown in green and blue boxes, respectively. CC = correlation coefficient; RV = relative variability; KGE = Kling‐Gupta Efficiency; PBIAS = percentage bias; BC = bias correction.CC BY · PubMed Central
Figure 10 Same as Figure 9, but for (a, left) drainage area and (b, right) aridity.
Figure 10 Same as Figure 9, but for (a, left) drainage area and (b, right) aridity.CC BY · PubMed Central
Figure 11 (a) The global river width and (b) the Variable Infiltration Capacity‐Routing Application for Parallel computatIon of Discharge‐simulated mean annual discharge for the river reaches observable by SWOT (width >50 m, as determined using GRWL)—there are 211,968 such reaches, taking up 8.8% of the total modeled reaches. Cross‐sectional width measurements are obtained from the GRWL database (Allen & Pavelsky, 2018) and are averaged over the Multi‐Error‐Removed‐Improved‐Terrain reaches, excluding measurements with lake/reservoir tags and multichannels. Note that SWOT will not observe a handful of reaches north of 78N, but they are not differentiated here. GRWL = Global River Widths from Landsat; SWOT = Surface Water and Ocean Topography.
Figure 11 (a) The global river width and (b) the Variable Infiltration Capacity‐Routing Application for Parallel computatIon of Discharge‐simulated mean annual discharge for the river reaches observable by SWOT (width >50 m, as determined using GRWL)—there are 211,968 such reaches, taking up 8.8% of the total modeled reaches. Cross‐sectional width measurements are obtained from the GRWL database (Allen & Pavelsky, 2018) and are averaged over the Multi‐Error‐Removed‐Improved‐Terrain reaches, excluding measurements with lake/reservoir tags and multichannels. Note that SWOT will not observe a handful of reaches north of 78N, but they are not differentiated here. GRWL = Global River Widths from Landsat; SWOT = Surface Water and Ocean Topography.CC BY · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 38 Not checked · 77 Own results and interpretations

62 more claims

Create a free account to see all 117 claims and to verify the 42 references still open.

See all claims — free account
  • Probably not a substantial claim. 24
  • It may state two separate claims as one. 30

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

13 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

1 of 1

Citations

Cites

Show all 80 references

Research works printed but not linked: 1

  1. Lehner B., 2006, HydroSHEDS technical documentation, version 1.0, 1

Cited by

Items 401–500 of 513

Metadata

Authors
Lin, P., Pan, M., Beck, H. E., Yang, Y., Yamazaki, D., Frasson, R., David, C. H., Durand, M., Pavelsky, T. M., Allen, G. H., Gleason, C. J., & Wood, E. F.
Affiliations
  • Princeton University (US)
  • Institute of Industrial Science, The University of Tokyo (JP)
  • The Ohio State University (US)
  • Jet Propulsion Laboratory (US)
  • University of North Carolina at Chapel Hill (US)
  • Texas A&M University (US)
  • University of Massachusetts Amherst (US)
Corresponding author
  • Lin, Peirong
  • Pan, Ming
Venue
Water Resources Research
Published
2019-08
Volume
55
Issue
8
Pages
6499-6516
ISSN
  • 0043-1397 (ISSN-L)
  • 1944-7973
OpenAlex
W2963255299
Cited by
511
References
80
Citations by year
2026: 81 · 2025: 70 · 2024: 81 · 2023: 79 · 2022: 104 · 2021: 60 · 2020: 30 · 2019: 5
FWCI
18.14
Type
journal article
Language
en
Open access
Locations
  • Water Resources Research (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • Carolina Digital Repository (University of North Carolina at Chapel Hill) (submittedVersion)
  • UNC Libraries
Funders
  • U.S. Army Corps of Engineers
  • Jet Propulsion Laboratory
  • National Aeronautics and Space Administration (NNX15AH05A)
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 › Environmental Engineering › Hydrological Forecasting Using AI
Keywords
Variable Infiltration Capacity model, river routing, runoff characteristics, machine learning calibration, bias correction, precipitation forcing, global digital elevation models, stream order

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).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

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

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.