Meander Morphology Classifier

Live demo

This project has a one-click Streamlit demo using public-safe synthetic or example data: Launch demo.

Problem

River meander planforms contain information about bend asymmetry, compound-bend structure and morphodynamic behaviour. Classification workflows need to connect centreline geometry, curvature spectra, machine-learning representations and reproducible outputs.

Approach

• Extract inflection-bounded single bends and compound/complex meander units from centreline data.
• Represent curvature structure using continuous wavelet-transform spectra.
• Use autoencoder latent spaces and clustering to support morphology classification.
• Provide command-line workflows and a Streamlit GUI for repeatable analysis and inspection.

What it demonstrates

This project shows scientific machine learning in a research-software setting: spectral features, autoencoder-based representation learning, latent-space interpretation, Streamlit tooling, a peer-reviewed methodology, documentation and reproducible workflows.

Publication

The classification framework behind this toolkit is described in a peer-reviewed paper:

Lopez Dubon, S., Sgarabotto, A., & Lanzoni, S. (2025). A Curvature-Based Framework for Automated Classification of Meander Bends. Water Resources Research, 61(2), e2024WR037583.