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

The DOI matches its Crossref record; 94 of the 94 references it cites were found in a registry or an index; none of the 47 sources its claims cite is checked for support yet.

Enhancing River Channel Dimension Estimation: A Machine Learning Approach Leveraging the National Water Model, Hydrographic Networks, and Landscape Characteristics

Open access
Modaresi Rad, Arash, Johnson, J. Michael, Ghahremani, Zahra, Coll, James, Frazier, Nels · 2024 · Journal of Geophysical Research: Machine Learning and Computation

Abstract

Knowledge of bankfull hydraulic geometry represents an essential requirement for various applications, including accurate flood prediction, hydrological routing, river behavior analysis, river management and engineering practices, water resource management, and beyond. Our work builds upon an extensive body of literature about estimating bankfull top‐width and depth at ungauged locations to enhance the understanding of observable factors that affect these parameters. Using more than 200,000 USGS Acoustic Doppler Current Profiler (ADCP) records, we developed a method employing machine learning (ML) using discharge estimates and landscape characteristics from sources, including the National Water Model (NWM), the National Hydrologic Geospatial Fabric network (NHGF), the EPA stream characteristic data set (StreamCat), and an array of satellite and reanalysis data products. Our method achieved log‐transformed R2 = 0.8 predicting bankfull depth (R2 = 0.77 for in‐channel conditions) and R2 = 0.76 predicting bankfull top‐width (R2 = 0.66 for in‐channel conditions) in the testing data set. The depth and width predictions showed lowest skill in mountainous and plateau regions. Our analysis demonstrates the benefit of data‐driven modeling in contrast to other global scaling‐based or regional statistical methods. In summary, our study illustrates how top‐width and depth can be better predicted using ML, reanalysis streamflow simulations, hydrographic networks, and summarized geospatial data.

Overview

CoChat Summary

TL;DR

The paper develops machine-learning models to estimate bankfull and in-channel river top-width and depth across CONUS using >200,000 USGS ADCP records and predictors from the National Water Model, hydrographic networks, and landscape characteristics. The meta-learner ensemble achieved log-transformed R² = 0.80 for bankfull depth and 0.76 for bankfull top-width, outperforming global/regional equations and a prior ML approach.

Key findings

  • Using >200,000 ADCP records from 10,081 USGS gauging locations, after filtering 4,229 of 6,226 initial sites were retained; testing-data log-transformed R² was 0.80 for bankfull depth (0.77 in-channel) and 0.76 for bankfull top-width (0.66 in-channel).
  • The most influential predictor was bankfull/in-channel discharge, especially NWM-derived flood-frequency principal components; width was also influenced by cumulative upstream flowline length (Arbolatesu), topographic wetness index, and actual evapotranspiration, while depth was influenced by mean catchment elevation, channel slope, and base flow index.
  • Model skill varied by hydrologic landscape region: depth models performed better in plains and permeable-bedrock regions, while top-width models had lowest GOF in HLR 2 (very humid plains, R² ≈ 0.53) and highest in mountainous/plateau regions; the abstract reports lowest skill in mountainous and plateau regions.
  • The ML models outperformed global drainage-area and discharge-based equations and the Doyle et al. (2023) ML method, with reduced scatter for mid-size and larger rivers but overprediction in smaller rivers.

Methods

  • Data: HYDRoSWOT ADCP observations (200,000+ measurements at 10,081 USGS gauges); profiles filtered to 2000–2015, depth <65 m, top-width <4 km, discharge agreement with NWIS, and no negative values; sites with inverse discharge–depth relationships or <5 profiles removed; AHG fits retained 4,229 of 6,226 sites with R² > 0.6. Bankfull was defined as 50% annual exceedance probability discharge and in-channel as 100%, computed via flood-frequency distributions (GEV, GP, LP3, generalized gamma; KS selection) and AHG relations.
  • Predictors: 116 initial variables from NWM v2.1 retrospective streamflow statistics (20%–0.1% annual exceedance probability and flow percentiles), NHGF hydrographic network, EPA StreamCat (58 watershed predictors), and climateR-aggregated soil/landscape/weather data (31 predictors); reduced to 15 features using SHAP and PCA/correlation clustering.
  • Modeling: trained 40 ML models (neural network, tree-based, SVM), selected top 10 by R²/RMSE, fine-tuned with five-fold cross-validation and three repeats, then built meta-learner and voting ensembles; used 70% training, 10% validation, 20% testing; applied to approximately 2.7 million NHGF reaches across CONUS.

Limitations

  • Paper states: training data are constrained to USGS sites, often with high banks, and may not represent entire reach geometry; no ADCP measurements in estuaries near river mouths make estimates unreliable there; estimates are unreliable near anthropogenically altered systems (dams, levees, irrigation canals, ponds); NWM’s diminished skill for 100% annual exceedance probability discharge impacts model GOF; reach-scale resolution overlooks braiding, incision, and aggradation; global scalability is constrained by reliance on NWM retrospective data and source-data biases/inconsistencies.
  • Text makes evident: models underpredict top-width and depth for smaller rivers, and there is a shortage of large-river data in HYDRoSWOT; performance declines with lower discharge; top-width model GOF is lowest in HLR 2 plains, while depth model GOF is lower in mountainous/impermeable-bedrock regions.

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

Integrity signals

Registry

Crossref · 10.1029/2024jh000173

Cites

94 of 94 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Comparison of NWM and NWIS flood discharges. (a) In-channel or 100% annual exceedance probability flow comparison; and (b) bankfull or 50% annual exceedance probability comparison.
Figure 1 Comparison of NWM and NWIS flood discharges. (a) In-channel or 100% annual exceedance probability flow comparison; and (b) bankfull or 50% annual exceedance probability comparison.CC-BY · From the PDF
Image 2
CC-BY · From the PDF
Figure 2 Schematic representation of model training for in-channel and bankfull flow top-width and depth.
Figure 2 Schematic representation of model training for in-channel and bankfull flow top-width and depth.CC-BY · From the PDF
Image 4
CC-BY · From the PDF
Figure 3 Predicted bankfull depth and top-width for all used stations in HYDRoSWOT data mapped to reference fabric flowlines using stream segment IDs (COMID). Using meta-learner (subplots a and b, respectively). Subplots (c) and (d) represent the meta-learner channel depth GOF for in-channel and bankfull flow conditions. Subplots e and f represent the meta-learner channel top-width GOF for in-channel and bankfull flow conditions.
Figure 3 Predicted bankfull depth and top-width for all used stations in HYDRoSWOT data mapped to reference fabric flowlines using stream segment IDs (COMID). Using meta-learner (subplots a and b, respectively). Subplots (c) and (d) represent the meta-learner channel depth GOF for in-channel and bankfull flow conditions. Subplots e and f represent the meta-learner channel top-width GOF for in-channel and bankfull flow conditions.CC-BY · From the PDF
Image 6
CC-BY · From the PDF
Figure 4 SHAP features importance values, distribution, and impact on model prediction for bankfull top-width (a) and depth (b). All NWM Flood principal components (PCs) are uncorrelated representations (PCs are orthogonal to each other) of 20%-0.1% annual exceedance probability discharge. Human PC2 is a representation of the total reservoir volume in the catchment. Soil PC0 represents soil moisture, Soil PC1 represents % of clay and silt, and Soil PC2 represents % of sand.
Figure 4 SHAP features importance values, distribution, and impact on model prediction for bankfull top-width (a) and depth (b). All NWM Flood principal components (PCs) are uncorrelated representations (PCs are orthogonal to each other) of 20%-0.1% annual exceedance probability discharge. Human PC2 is a representation of the total reservoir volume in the catchment. Soil PC0 represents soil moisture, Soil PC1 represents % of clay and silt, and Soil PC2 represents % of sand.CC-BY · From the PDF
Image 8
CC-BY · From the PDF
Figure 5 MLmodel performance across different quantiles of the most influential variables for both depth and width models. Within each subplot each dot represents a station belonging to a quantile of an important feature (a)-(d) belong to grouped performances of top-width and subplots (e)-(h) belong to depth. Th model skill ( R 2 ) is shown as cyan plus sign.
Figure 5 MLmodel performance across different quantiles of the most influential variables for both depth and width models. Within each subplot each dot represents a station belonging to a quantile of an important feature (a)-(d) belong to grouped performances of top-width and subplots (e)-(h) belong to depth. Th model skill ( R 2 ) is shown as cyan plus sign.CC-BY · From the PDF
Image 10
CC-BY · From the PDF
Figure 6 Representation of the contribution of each important feature and its inner correlation and impact on the top-width and depth models. (a) represents the impact of the topological wetness index on the channel top-width ML model and its correlation with the NWM 50% annual exceedance probability discharge. (b) represents the impact of AET on the channel depth ML model and its correlation with % of soil clay and silt content.
Figure 6 Representation of the contribution of each important feature and its inner correlation and impact on the top-width and depth models. (a) represents the impact of the topological wetness index on the channel top-width ML model and its correlation with the NWM 50% annual exceedance probability discharge. (b) represents the impact of AET on the channel depth ML model and its correlation with % of soil clay and silt content.CC-BY · From the PDF
Figure 7 Depth ML model GOF ( R 2 ) comparison with literature and across hydrologic landscape regions (a). (b) the GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived depth from NWIS bankfull depth measurements. (c) GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived depth from NWIS in-channel depth measurements. (d) shows the population of sites and distribution of depth per HLR in HYDRoSWOT. (e) proposed ML model GOF ( R 2 ) comparison (in columns) to Blackburn-Lynch et al. (2017) (in black dots) across hydrologic landscape regions.
Figure 7 Depth ML model GOF ( R 2 ) comparison with literature and across hydrologic landscape regions (a). (b) the GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived depth from NWIS bankfull depth measurements. (c) GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived depth from NWIS in-channel depth measurements. (d) shows the population of sites and distribution of depth per HLR in HYDRoSWOT. (e) proposed ML model GOF ( R 2 ) comparison (in columns) to Blackburn-Lynch et al. (2017) (in black dots) across hydrologic landscape regions.CC-BY · From the PDF
Image 13
CC-BY · From the PDF
Figure 8 Width ML model GOF ( R 2 ) comparison with literature and across hydrologic landscape regions (a). (b) the GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived width from NWIS bankfull width measurements. (c) GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived width from NWIS in-channel width measurements. (d) shows the population of sites and distribution of width per HLR in HYDRoSWOT (e) proposed ML model GOF ( R 2 ) comparison (in columns) to Blackburn-Lynch et al. (2017) (in black dots) across hydrologic landscape regions.
Figure 8 Width ML model GOF ( R 2 ) comparison with literature and across hydrologic landscape regions (a). (b) the GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived width from NWIS bankfull width measurements. (c) GOF ( R 2 ) of Blackburn-Lynch et al. (2017) derived width from NWIS in-channel width measurements. (d) shows the population of sites and distribution of width per HLR in HYDRoSWOT (e) proposed ML model GOF ( R 2 ) comparison (in columns) to Blackburn-Lynch et al. (2017) (in black dots) across hydrologic landscape regions.CC-BY · From the PDF
Image 15
CC-BY · From the PDF
Figure 9 Channel top-width at bankfull and in-channel flow conditions model performance comparison between (a, e) global drainage area-based equations, (b, f) global discharge-based equations, (c, g) the ML model proposed by Doyle et al. (2023), and (d, h) our proposed model.
Figure 9 Channel top-width at bankfull and in-channel flow conditions model performance comparison between (a, e) global drainage area-based equations, (b, f) global discharge-based equations, (c, g) the ML model proposed by Doyle et al. (2023), and (d, h) our proposed model.CC-BY · From the PDF
Image 17
CC-BY · From the PDF
Figure 10 Channel depth at bankfull and in-channel flow conditions model performance comparison between (a, e) global drainage area-based equations, (b, f) global discharge-based equations, (c, g) the ML model proposed by Doyle et al. (2023), and (d, h) our proposed model.
Figure 10 Channel depth at bankfull and in-channel flow conditions model performance comparison between (a, e) global drainage area-based equations, (b, f) global discharge-based equations, (c, g) the ML model proposed by Doyle et al. (2023), and (d, h) our proposed model.CC-BY · From the PDF
Image 19
CC-BY · From the PDF

Verification

Claims the paper makes

0 Valid · 1 Not valid · 30 Not checked · 79 Own results and interpretations

57 more claims

Create a free account to see all 110 claims and to verify the 35 references still open.

See all claims — free account
  • Probably not a substantial claim. 22
  • It may state two separate claims as one. 33
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

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

Each contradiction between two passages listed here was confirmed by an independent check.

4 of 4

Citations

Cites

Show all 94 references

Research works printed but not linked: 3

  1. Bock A. R., 2022, Provisional: National hydrologic geospatial fabric reference and derived hydrofabrics, 2022120390
  2. Haan C. T. Johnson H. P. &Brakensiek D. L.(1982).Hydrologic modeling of small watersheds.https://doi.org/10.5555/19851999449 · doi:10.5555/19851999449
  3. Köppen M., 2000, 5th Online World Conference on Soft Computing in Industrial Applications (WSC5), 4

Cited by

See all 11

Metadata

Authors
Modaresi Rad, A., Johnson, J. M., Ghahremani, Z., Coll, J., & Frazier, N.
Affiliations
  • Lynker Fort Collins CO USA
  • National Oceanic and Atmospheric Administration (US)
Corresponding author
Johnson, J. Michael
Venue
Journal of Geophysical Research: Machine Learning and Computation
Published
2024-11-25
Volume
1
Issue
4
ISSN
2993-5210 (ISSN-L)
OpenAlex
W4404699222
Cited by
11
References
94
Citations by year
2026: 7 · 2025: 4
FWCI
2.26
Type
journal article
Language
en
Open access
Locations
  • Journal of Geophysical Research Machine Learning and Computation (publishedVersion, cc-by)
  • NOAA Institutional Repository (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Oceanic and Atmospheric Administration
Topics
  • Physical Sciences › Earth and Planetary Sciences › Earth-Surface Processes › Hydrology and Sediment Transport Processes
  • Physical Sciences › Environmental Science › Water Science and Technology › Hydrology and Watershed Management Studies
  • Physical Sciences › Environmental Science › Environmental Engineering › Hydrological Forecasting Using AI
Keywords
bankfull depth, ungauged rivers, National Water Model, river networks, landscape elements, acoustic Doppler current profiler, discharge estimation, flood forecasting

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: Enhancing River Channel Dimension Estimation: A Machine Learning Approach Leveraging the National Water Model, Hydrographic Networks, and Landscape Characteristics. https://app.cochat.ai/papers/10.1029/2024jh000173
The paper
Modaresi Rad, A., Johnson, J. M., Ghahremani, Z., Coll, J., & Frazier, N. (2024). Enhancing River Channel Dimension Estimation: A Machine Learning Approach Leveraging the National Water Model, Hydrographic Networks, and Landscape Characteristics. Journal of Geophysical Research: Machine Learning and Computation. https://doi.org/10.1029/2024jh000173

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.