The Southeastern United States rivers meander classification based on geometric properties such as sinuosity, wavelength, length, amplitude, width and Normalized Meander Curvature Index (NMCI).This study investigates the influence of geometric properties and physiographic regions on river meander classification in the southeastern United States. We employed four machine learning models—Gradient Boosting Machine, Neural Network, Decision Tree, and Random Forest—to classify meanders based on their amplitude, wavelength, sinuosity, water-surface width, and Normalized Meander Curvature Index (NMCI).
- Gradient Boosting Machine
- Neural Network
- Decision Tree
- Random Forest
- Amplitude
- Wavelength
- Sinuosity
- Water-surface width
- Normalized Meander Curvature Index (NMCI)
- Analysis of Variance (ANOVA)
- Kruskal-Wallis tests
- Tukey’s HSD post-hoc comparisons
Our statistical analyses revealed significant differences in geometric properties among meander classes and across physiographic regions (Coastal Plain, Piedmont, Appalachian Plateau). Among the machine learning models, the Random Forest model demonstrated the highest classification accuracy at 85%, outperforming the Decision Tree (82%), Neural Network (77%), and Gradient Boosting Machine (67%). Zenode:https://zenodo.org/records/15730646?preview=1
The source code is hosted on GitHub at https://github.com/thapawan/MeanderClassification. Moreover, GEE apps are hosted here: https://meanderclassify.users.earthengine.app/view/slopecurvature. The centerline and water-surface width estimated from MAT are shared in the Google Earth Engine Apps: https://meanderclassify.users.earthengine.app/view/matcw.
Open for collaboration and welcome any valuable feedback or suggestions for improvement. If you have any queries about the algorithm, open for discussion and contact: pthapa2@crimson.ua.edu.
Thapa P. (2025) Geospatial Classification of River Meanders: A Machine Learning Framework Driven by Deep Learning and Geometric Properties.