A remote sensing-based approach to suspended sediment estimation in rivers with scarce in situ data
Campos, Juliana Andrade LU
; Uvo, Cintia Bertacchi
LU
; Fujita, Thais
LU
; Cuartas, Luz Adriana
LU
and Monteiro Pontes, Paulo Rógenes
(2026)
In Journal of Hydrology
677.
- Abstract
Monitoring suspended sediment concentration (SSC) is essential for water-quality management and for anticipating sediment-related impacts. Yet sparse in situ measurements limit tracking SSC at high temporal and spatial resolution. We address this gap with a scalable approach that couples multispectral imagery (Landsat-8/9, Sentinel-2), remote sensing precipitation product (CHIRPS), and soil properties product (SoilGrids) within artificial neural networks (ANNs) to estimate SSC in the Doce and Upper São Francisco River basins, Brazil. Data acquisition and preprocessing were performed in Google Earth Engine, and ANNs were trained in MATLAB and ported to GEE for pixel-wise mapping. We compiled 978 SSC observations from 41 gauging stations... (More)
Monitoring suspended sediment concentration (SSC) is essential for water-quality management and for anticipating sediment-related impacts. Yet sparse in situ measurements limit tracking SSC at high temporal and spatial resolution. We address this gap with a scalable approach that couples multispectral imagery (Landsat-8/9, Sentinel-2), remote sensing precipitation product (CHIRPS), and soil properties product (SoilGrids) within artificial neural networks (ANNs) to estimate SSC in the Doce and Upper São Francisco River basins, Brazil. Data acquisition and preprocessing were performed in Google Earth Engine, and ANNs were trained in MATLAB and ported to GEE for pixel-wise mapping. We compiled 978 SSC observations from 41 gauging stations (rivers >30 m wide) to train and validate three model families: reflectance-only, precipitation/soil-only, and an integrated model. Reflectance-based ANNs best captured peaks and event-driven anomalies, including non-rainfall disturbances, while the precipitation/soil model produced stable daily estimates during cloudy periods, reproducing seasonal cycles. The integrated model reconciled both controls and achieved the best overall skill (R2 up to 0.84; MAE ≈ 18.7 mg L−1). The results show that precipitation, soil properties, and reflectance are complementary predictors that can be combined to deliver continuous, spatially explicit SSC information in data-scarce basins. The workflow is cost-effective, reproducible, and readily deployable for routine monitoring, post-event assessment, and planning of erosion control and reservoir siltation management.
(Less)
- author
- Campos, Juliana Andrade
LU
; Uvo, Cintia Bertacchi
LU
; Fujita, Thais
LU
; Cuartas, Luz Adriana
LU
and Monteiro Pontes, Paulo Rógenes
- organization
- publishing date
- 2026-09
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Artificial neural networks, Data scarcity, Landsat, Sediment transport, Sentinel-2
- in
- Journal of Hydrology
- volume
- 677
- article number
- 135779
- publisher
- Elsevier
- external identifiers
-
- scopus:105042111982
- ISSN
- 0022-1694
- DOI
- 10.1016/j.jhydrol.2026.135779
- language
- English
- LU publication?
- yes
- id
- 6f3a8bb0-0160-4639-98a0-472192aed6cd
- date added to LUP
- 2026-07-03 09:42:29
- date last changed
- 2026-07-03 09:43:35
@article{6f3a8bb0-0160-4639-98a0-472192aed6cd,
abstract = {{<p>Monitoring suspended sediment concentration (SSC) is essential for water-quality management and for anticipating sediment-related impacts. Yet sparse in situ measurements limit tracking SSC at high temporal and spatial resolution. We address this gap with a scalable approach that couples multispectral imagery (Landsat-8/9, Sentinel-2), remote sensing precipitation product (CHIRPS), and soil properties product (SoilGrids) within artificial neural networks (ANNs) to estimate SSC in the Doce and Upper São Francisco River basins, Brazil. Data acquisition and preprocessing were performed in Google Earth Engine, and ANNs were trained in MATLAB and ported to GEE for pixel-wise mapping. We compiled 978 SSC observations from 41 gauging stations (rivers >30 m wide) to train and validate three model families: reflectance-only, precipitation/soil-only, and an integrated model. Reflectance-based ANNs best captured peaks and event-driven anomalies, including non-rainfall disturbances, while the precipitation/soil model produced stable daily estimates during cloudy periods, reproducing seasonal cycles. The integrated model reconciled both controls and achieved the best overall skill (R<sup>2</sup> up to 0.84; MAE ≈ 18.7 mg L<sup>−1</sup>). The results show that precipitation, soil properties, and reflectance are complementary predictors that can be combined to deliver continuous, spatially explicit SSC information in data-scarce basins. The workflow is cost-effective, reproducible, and readily deployable for routine monitoring, post-event assessment, and planning of erosion control and reservoir siltation management.</p>}},
author = {{Campos, Juliana Andrade and Uvo, Cintia Bertacchi and Fujita, Thais and Cuartas, Luz Adriana and Monteiro Pontes, Paulo Rógenes}},
issn = {{0022-1694}},
keywords = {{Artificial neural networks; Data scarcity; Landsat; Sediment transport; Sentinel-2}},
language = {{eng}},
publisher = {{Elsevier}},
series = {{Journal of Hydrology}},
title = {{A remote sensing-based approach to suspended sediment estimation in rivers with scarce in situ data}},
url = {{http://dx.doi.org/10.1016/j.jhydrol.2026.135779}},
doi = {{10.1016/j.jhydrol.2026.135779}},
volume = {{677}},
year = {{2026}},
}