CoChat
Sign Up
Cited by 156Hydrology and Sediment Transport ProcessesFlood Risk Assessment and ManagementRemote-Sensing Image Classification

The DOI matches its Crossref record; 28 of the 28 references it cites were found in a registry or an index; none of the 33 sources its claims cite is checked for support yet.

RivWidthCloud: An Automated Google Earth Engine Algorithm for River Width Extraction From Remotely Sensed Imagery

Open access
Yang, Xiao, Pavelsky, Tamlin M., Allen, George H., Donchyts, Gennadii·2020·IEEE Geoscience and Remote Sensing Letters
DOI 10.1109/lgrs.2019.2920225
Open this paper in CoChat — free account

Abstract

The wetted width of a river is one of the most important hydraulic parameters that can be readily measured using remote sensing. Remotely sensed river widths are used to estimate key attributes of river systems, including changes in their surface area, channel storage, and discharge. Although several published algorithms automate river network and width extraction from remote sensing images, they are limited by only being able to run on local computers and do not automatically manage cloudy images as input. Here we present RivWidthCloud, a river width software package developed on the Google Earth Engine cloud computing platform. RivWidthCloud automatically extracts river centerline and widths from optical satellite images with the ability to flag observations that are obstructed by features like clouds, cloud shadows, and snow based on existing quality band classification. Because RivWidthCloud is built on a popular cloud computing platform, it allows users to easily apply the algorithm to the platform’s vast archive of remote sensing images, thereby reducing the users’ overhead for computing hardware and data storage. By comparing RivWidthCloud-derived widths from Landsat images to in situ widths from the U.S. and Canada, we show that RivWidthCloud can estimate widths with high accuracy (root mean square error: 99 m; mean absolute error: 43 m; mean bias: −21 m). By making RivWidthCloud publicly available, we anticipate that it will be used to address both river science questions and operational applications of water resource management.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

RivWidthCloud is a Google Earth Engine software package that automatically extracts river centerlines and widths from optical satellite imagery, including partially cloudy images, by adapting RivWidth and adding quality flags. Validation against same-day in situ widths from U.S. and Canada Landsat data shows high accuracy (RMSE 99.2 m; MAE 43.1 m; mean bias −20 m).

Key findings

  • RivWidthCloud extracts river centerlines and widths in three main steps: river mask, centerline, and width measurement; it uses GRWL to distinguish river pixels from nonriver water, checks connectivity to GRWL centerlines within a default 4-km radius, and removes islands smaller than a default ∼0.3 km².
  • It provides quality flags for cloud, cloud shadow, snow/ice, and topographic shadow; a nonzero flag suggests the width measurement may be impacted by that condition.
  • Validation used 1514 valid flag-free widths from Landsat images with same-day in situ measurements for 1984–2018, across the U.S. (492 stations with 1476 widths) and Canada (27 stations with 38 widths); RMSE was 99.2 m, MAE 43.1 m, and mean bias −20 m.
  • Unlike RivWidth, RivWidthCloud preserves each channel’s own centerline and measures each channel’s width separately rather than outputting a single centerline with total flow width summed across all channels.

Methods

  • Design: a Google Earth Engine implementation of RivWidth that works in principle with any remote sensing product providing water and obstruction classifications, but is described and validated using Landsat TM, ETM+, and OLI imagery.
  • River mask: spectral water classification with reported ∼97% accuracy using MNDWI, NDVI, and EVI thresholds; channel mask is all water pixels connected to GRWL river centerlines within a default 4-km radius; islands below ∼0.3 km² are removed.
  • Centerline and width: distance transform, gradient convolution, thresholding gradient values <0.9, and skeletonization produce a 1-pixel-wide centerline; spurious branches are trimmed by removing 500 pixels (∼1500 m for Landsat) or the segment length to the next confluence, whichever is shorter; widths are calculated along orthogonal directions using a 9 × 9-pixel kernel and w = w_m × u. Validation compares Landsat-derived widths to USGS and Water Survey of Canada same-day in situ widths.

Limitations

  • Stated: default centerlines do not guarantee connectivity at confluences; connectivity can be enforced by calculating the centerline from the river mask via skeletonization only, but this is slow and could exceed available GEE computational resources over large areas.
  • Stated: width accuracy depends on water classification and on cloud/shadow/snow/ice flagging accuracy; the error assessment is likely conservative because it does not account for in situ measurement errors or mismatches between in situ and satellite measurement locations where width changes rapidly.
  • Evident from the text: validation is limited to U.S. and Canada gauging stations and Landsat imagery, and RivWidthCloud is currently implemented only for Landsat products, though the authors state it can be expanded to other products such as Sentinel-2 or MODIS.

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

Integrity signals

Registry

Crossref · 10.1109/lgrs.2019.2920225

Cites

28 of 28 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 33 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    For the period 1984-2018, 1514 valid (flag-free) widths were obtained from Landsat images that had same-day in situ measurements at gauging stations, across the U.S. (492 stations with 1476 widths) and Canada (27 stations with 38 widths).

    “For the period 1984-2018, we obtained 1514 valid (flag-free) widths from Landsat images that had same-day in situ measurements at gauging stations, across the U.S. (492 stations with 1476 widths) and Canada (27 stations with 38 widths).”
    Quote in the paperSubstantial claim
  • Findingown data

    RivWidthCloud-derived widths from Landsat images closely matched same-day in situ width measurements from the U.S. and Canada, with root mean square error about 99 m, mean absolute error about 43 m, and mean bias about -20 m.

    “By comparing RivWidthCloudderived widths from Landsat images to in situ widths from the U.S. and Canada, we show that RivWidthCloud can estimate widths with high accuracy (root mean square error: 99 m; mean absolute error: 43 m; mean bias: -21 m).”
    • Also stated in: § III. RESULTS AND DISCUSSION · plain
      “We found that the widths computed using RivWidthCloud on Landsat imagery (w L ) closely matched the in situ measurements (w G ) (root mean square error = 99.2m; mean absolute error = 43.1 m; mean bias = -20 m) (see also Fig. 3).”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The resulting 1-pixel-wide centerlines often exhibit spurious branches, artifacts mainly produced by irregular bank shapes.

    “The resulting 1-pixel-wide centerlines often exhibit spurious branches-artifacts mainly produced by irregular bank shapes.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The centerlines from RivWidthCloud by default do not guarantee connectivity at confluences.

    “The centerlines from RivWidthCloud by default do not guarantee connectivity at confluences.”
    Quote in the paperSubstantial claim
  • Interpretationnone given

    The resulting orthogonal direction is only accurate when exactly two pixels intersect the rim of the kernel, one from each end of the overlapping centerline segment.

    “However, the resulting orthogonal direction is only accurate when exactly two pixels intersect the rim of the kernel, one from each end of the overlapping centerline segment (Fig. 2).”
    Quote in the paperSubstantial claim
  • It may state two separate claims as one. 13
  • Probably not a substantial claim. 10

  • It may state two separate claims as one.

    Findingreasoning

    The final centerline from RivWidthCloud preserves the river's channel structure such that each channel retains its own centerline, and as a result, each channel's width is measured separately.

    “The resulting final centerline [Fig.1(h)] preserves the river's channel structure such that each channel retains its own centerline, and as a result, each channel's width is measured separately.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Large inconsistencies between in situ and remotely sensed widths were mainly from stations where nearby river widths changed rapidly or the in situ width was taken a certain distance upstream or downstream from the gauging station.

    “Large inconsistencies between widths obtained in situ and remotely were mainly from stations where: 1) nearby river widths changed rapidly or 2) the in situ width was taken a certain distance upstream or downstream from the gaging station.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Findingreasoning

    RivWidthCloud is currently implemented with remote sensing products from the Landsat program.

    “RivWidthCloud is currently implemented with remote sensing products from Landsat program, given its long mission history, wide application, and sophisticated cloud algorithm.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Landsat has a long mission history.

  • It may state two separate claims as one.

    Interpretationreasoning

    Two distinct advantages of RivWidthCloud over previous river width measuring algorithms are its ease of use based on a popular cloud computing environment and its flagging capability to automatically reduce adverse impact from features like cloud and shadows.

    “Two distinct advantages of RivWidthCloud over previous river width measuring algorithms [1], [2], [9]- [14] are its ease of use based on a popular cloud computing environment and its flagging capability to automatically reduce adverse impact from features like cloud and shadows.”
    Quote in the paperSubstantial claim0 of 8 sources checked
    • cites [1]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [2]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [10]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [11]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [12]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [13]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [14]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Interpretationreasoning

    For the majority of width validation pairs, the accuracy of Landsat-derived widths depends on the accuracy of the image classification: the classification for water determines the extent of the river while the classification of cloud/shadow/snow/ice determines the flagging accuracy.

    “For the majority of the width validation pairs, the accuracy of Landsat-derived widths depends on the accuracy of the image classification-the classification for water determines the extent of the river while the classification of cloud/shadow/snow/ice determines the flagging accuracy.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Variations in centerline and width along river channels have been used to infer the morphodynamics of rivers and river deltas.

    “V ARIATIONS in centerline and width along river channels have been used to infer the morphodynamics of rivers and river deltas [1]- [3].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [1]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [2]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [3]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.

    Interpretationnone given

    The accuracy of the channel/river mask and the widths derived from them is highly dependent on the accuracy of the inundation map, and hence on the water classification method used to generate the inundation map.

    “We note that the accuracy of the channel/river mask and the widths derived from them is highly dependent on the accuracy of the inundation map, and hence on the water classification method that is used to generate the inundation map.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationreasoning

    Because RivWidthCloud is built on a popular cloud computing platform, it allows users to easily apply the algorithm to the platform's vast archive of remote sensing images, reducing users' overhead for computing hardware and data storage.

    “Because RivWidthCloud is built on a popular cloud computing platform, it allows users to easily apply the algorithm to the platform's vast archive of remote sensing images, thereby reducing the users' overhead for computing hardware and data storage.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Premisecitation

    Recent work has pioneered the ability to distinguish clouds, cloud shadows, topographic shadows, and snow/ice using a machine learning approach.

    “Recent work has pioneered the ability to distinguish these features using a machine learning approach [25], [26]”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [25]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [26]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Interpretationnone given

    Solely relying on skeletonization to calculate the centerline is slow and could potentially exceed the computational resources available to a GEE user when applied over a large area.

    “However, solely relying on skeletonization to calculate the centerline is slow and could potentially exceed the computational resources available to a GEE user when applied over a large area.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim.

    Premisecitation

    At any given time, clouds cover approximately 67% of the earth's surface.

    “At any given time, clouds cover ∼67% of the earth's surface [15]”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim.

    Premisecitation

    Point-scale and reach-scale multitemporal river widths have been used to estimate river discharge using hydraulic geometry relationships established with in situ discharge measurements.

    “For example, point-scale and reach-scale [6] multitemporal river widths have been used to estimate river discharge using hydraulic geometry relationships established with in situ discharge measurements.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [6]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Interpretationreasoning

    RivWidthCloud in principle works with any remote sensing product capable of providing classifications of water and visual obstructions (cloud, snow/ice, etc.).

    “RivWidthCloud in principle works with any remote sensing product capable of providing classifications of water and visual obstructions (cloud, snow/ice, etc.).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The methods for classifying water and the accuracy of such classifications vary among remote sensing products.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Interpretationreasoning

    A flag with a nonzero value suggests that the width measurement may be impacted by the flagged condition.

    “Thus, a flag with a nonzero value suggests the width measurement may be impacted by the flagged condition.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationreasoning

    In combination with increasing earth-observing satellite missions, including SWOT, satellite-derived widths will help observe ungauged rivers and supplement in situ gauging networks, ensuring greater monitoring consistency globally.

    “In combination with the increasing number of earth-observing satellite missions, including the forthcoming Surface Water and Ocean Topography (SWOT) mission [8], satellite-derived widths will help us observe ungauged rivers and supplement in situ gauging networks, ensuring greater monitoring consistency globally.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [8]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

    Interpretationnone given

    The range of remote sensing data for RivWidthCloud can be readily expanded to include other remote sensing data products (e.g., from Sentinel 2 or Moderate Resolution Imaging Spectroradiometer).

    “However, the range of remote sensing data for RivWidthCloud can be readily expanded to include other remote sensing data products (e.g., from Sentinel 2 or Moderate Resolution Imaging Spectroradiometer).”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationreasoning

    The accuracy of RivWidthCloud's width measurements depends on the accuracy of the water classification, which can be impacted by clouds, cloud shadows, topographic shadows, and the presence of snow/ice.

    “The accuracy of RivWidthCloud's width measurements depends on the accuracy of the water classification, which can be impacted by clouds, cloud shadows, topographic shadows, and the presence of snow/ice.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Premisecitation

    Most existing algorithms to extract river widths are designed for optical images and are heavily limited by clouds because they require cloud-free images as input.

    “Most existing algorithms to extract river widths are designed for optical images, and their application is heavily limited by the presence of clouds, as they require cloud-free images as input [1], [2], [9]- [14].”
    Quote in the paperSubstantial claim0 of 8 sources checked
    • cites [1]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [2]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [10]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [11]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [12]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [13]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [14]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • It may state two separate claims as one.

    Interpretationreasoning

    The gradient close to the centerline is nearly always a local minimum and close to zero, while the maximum distance scales with river size.

    “While the maximum distance scales with river size, the gradient close to the centerline is nearly always a local minimum and close to zero [Fig.1(f)].”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Interpretationreasoning

    RivWidthCloud allows for the generation of spatially continuous multitemporal width data that can be used to provide a priori data set of variations of river location and width for future satellite missions like SWOT, or to estimate river discharge either via AMHG alone or in conjunction with other remote sensing data products.

    “It also allows for the generation of spatially continuous multitemporal width data that can be used to provide a priori data set of variations of river location and width for future satellite missions like SWOT, or to estimate river discharge either via AMHG alone or via in conjunction with other remote sensing data products.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Interpretationreasoning

    In cases where connectivity is needed, it can be enforced by directly calculating the centerline from the river mask via skeletonization only, skipping the steps of distance transform and gradient.

    “In cases where connectivity is needed, it can be enforced by directly calculating the centerline from the river mask via skeletonization only, skipping the steps of distance transform and gradient.”
    Quote in the paperNot a substantial claim

  • It may state two separate claims as one.

    Interpretationreasoning

    RivWidthCloud automatically extracts river centerline and widths from optical satellite images and can flag observations obstructed by clouds, cloud shadows, and snow based on existing quality band classification.

    “RivWidthCloud automatically extracts river centerline and widths from optical satellite images with the ability to flag observations that are obstructed by features like clouds, cloud shadows, and snow based on existing quality band classification.”
    • Also stated in: § I. INTRODUCTION · plain
      “Here, we present RivWidthCloud, the first Google Earth Engine (GEE)-based river width algorithm that automatically extracts river centerlines and widths from remotely sensed images (cloud-free or not) with minimal user input and minimal required computation hardware.”
    Quote in the paperSubstantial claim
22 of 22

  • The paper relies on this without support.

    Many satellite-based river discharge estimation methods that use river observations require river width as input.

    “Among various satellite-based river discharge estimation methods that use river observations, many require river width as input.”

  • The paper relies on this without support.

    Landsat has a long mission history.

    “Because of its accurate cloud classification algorithm [17] and its long mission history, we describe and validate RivWidthCloud using data from the Landsat program, specifically imagery from the Thematic Mapper (TM), Enhanced TM Plus, and Operational Land Imager instruments.”

  • The paper relies on this without support.

    The methods for classifying water and the accuracy of such classifications vary among remote sensing products.

    “The methods for classifying water and the accuracy of such classifications vary among remote sensing products.”
3 of 3

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.

  • MERIT DEM citation points to a DSWE paper

    Minor

    The method cites reference 23 for MERIT DEM, but reference 23 is listed as a paper about Dynamic Surface Water Extent detection.

  • 500 Landsat pixels are about 15,000 m, not 1,500 m

    Minor

    At 30 m per Landsat pixel, 500 pixels span approximately 15,000 m. The stated 1,500 m is a factor of ten too small.

    Reported 1500 · recomputed 15000

  • Mean bias differs between the abstract and results

    Minor

    The abstract reports a mean bias of -21 m, while the results report -20 m for the same Landsat-versus-in-situ width validation.

3 of 3

Citations

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Metadata

Authors
Yang, X., Pavelsky, T. M., Allen, G. H., & Donchyts, G.
Affiliations
  • University of North Carolina at Chapel Hill (US)
  • Texas A&M University (US)
  • Deltares (NL)
ORCID
Venue
IEEE Geoscience and Remote Sensing Letters
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2020-02
Volume
17
Issue
2
Pages
217-221
ISSN
  • 1545-598X (ISSN-L)
  • 1558-0571
DOI
10.1109/lgrs.2019.2920225
OpenAlex
W2955998290
Cited by
156
References
26
Citations by year
2026: 31 · 2025: 26 · 2024: 21 · 2023: 18 · 2022: 16 · 2021: 24 · 2020: 19
FWCI
10.62
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Geoscience and Remote Sensing Letters (publishedVersion)
  • Carolina Digital Repository (University of North Carolina at Chapel Hill) (submittedVersion)
  • UNC Libraries
Funders
Jet Propulsion Laboratory
Topics
  • Physical Sciences › Earth and Planetary Sciences › Earth-Surface Processes › Hydrology and Sediment Transport Processes
  • Physical Sciences › Environmental Science › Global and Planetary Change › Flood Risk Assessment and Management
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Remote-Sensing Image Classification
Keywords
cloud detection, cloud shadows, Landsat imagery, Google Earth Engine, cloud computing, automated image processing, water resources management

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 2026-11-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-08 (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