Journal article
Enhancing river channel geometry estimation for bankfull conditions in Iowa using machine learning
Journal of hydrology (Amsterdam), Vol.665, 134753
02/2026
DOI: 10.1016/j.jhydrol.2025.134753
Abstract
•Accurate representation of stream channel geometry plays a critical role in hydraulic modeling and Flood Inundation Mapping.•Developing a methodology for estimating channel geometric characteristics over large spatial scales provides a cost-effective way to significantly improve synthetic rating curves and Flood Inundation Mapping techniques.•Machine learning techniques address gaps in river channel data, offering scalable and cost-effective solutions for hydrological and hydraulic modeling applications.
Accurate estimation of river channel geometry is critical for hydraulic modeling and Flood Inundation Mapping (FIM). However, some common data sources, such as Light Detection and Ranging (LiDAR), particularly Near-Infrared LiDAR (NIR), fail to capture underwater features, and regional bathymetric surveys are often cost-prohibitive. In this study, machine learning (ML) models were developed and tested to predict three geometric variables under bankfull conditions: top width, bottom width, and depth. The models used Random Forest Regression (RFR) and Nearest Neighbors Regression (k-NNR), leveraging independent variables derived from extensive geospatial datasets, including drainage area, stream order, slope, and sinuosity. Results demonstrated strong predictive performance for top and bottom width models and moderate performance for depth models. RFR slightly outperformed k-NNR in top width predictions, while k-NNR performed better in bottom width and depth estimations. Additional validation confirmed the models’ accuracy, closely aligning with testing results. The ML models produced significantly improved channel geometry estimates compared to LiDAR, reducing errors in depth and cross-sectional area for selected sites. This study highlights the potential of ML algorithms to enhance channel geometry representation, supporting the generation of synthetic rating curves and improving the accuracy of FIM. The findings suggest that combining RFR and k-NNR could further optimize predictions, offering a balance between robustness and precision for large-scale hydraulic modeling efforts.
Details
- Title: Subtitle
- Enhancing river channel geometry estimation for bankfull conditions in Iowa using machine learning
- Creators
- Marcela RojasDaniel GillesNathan Young
- Resource Type
- Journal article
- Publication Details
- Journal of hydrology (Amsterdam), Vol.665, 134753
- DOI
- 10.1016/j.jhydrol.2025.134753
- ISSN
- 0022-1694
- eISSN
- 1879-2707
- Publisher
- Elsevier B.V
- Grant note
- Iowa Flood Center
The authors acknowledge the support of the Iowa Flood Center, particularly our colleague Felipe Quintero, for valuable discussions and assistance in reviewing this document.
- Language
- English
- Date published
- 02/2026
- Academic Unit
- Civil and Environmental Engineering; IIHR--Hydroscience and Engineering
- Record Identifier
- 9985093884602771
Metrics
1 Record Views