Enhanced River Planform Analysis Using Deep Learning and Medial Axis Transform with Sentinel‑1A Imagery

APA Cite: Thapa, P., Davis, L., Amanambu, A., LaFevor, M., & Frame, J. (2025). Enhanced river planform analysis using deep learning and medial axis transform with Sentinel 1A imagery. Earth Surface Processes and Landforms50(12), e70158.


🌊 Introduction: Why Mapping Rivers Matters

Rivers are constantly changing — they migrate, widen, narrow, and shift across their floodplains. Tracking these changes is essential for:

  • Flood hazard planning
  • River restoration
  • Dam impact analysis
  • Ecological management
  • Understanding long‑term geomorphic evolution

However, accurately detecting river centerlines and water surface widths has been a persistent challenge, especially using satellite data.

Clouds, vegetation, watercolor, turbidity, and seasonal variability make optical imagery inconsistent.

This is where SAR (Synthetic Aperture Radar) and deep learning provide a major advantage.


🤖 A New Hybrid Method

Our study introduces a new approach that integrates:

1️⃣ DeepLabV3 – A State‑of‑the‑Art Deep Learning Model

DeepLabV3 is a powerful semantic segmentation model that can extract water bodies from Sentinel‑1A SAR imagery even when:

  • Clouds block optical sensors
  • The terrain is complex
  • Traditional thresholding fails

The model provides a clean and accurate water mask, which is the foundation for all subsequent measurements.


2️⃣ Medial Axis Transform (MAT) – Precision Geometry

Once the river boundaries are extracted, the Medial Axis Transform:

  • Finds the geometric centerline
  • Computes the distance to the riverbanks
  • Estimates water surface width
  • Preserves complex curvature information

This produces a continuous river centerline and precise width measurements — essential for analyzing river planform dynamics.


📌 Study Areas

We tested this approach in multiple river systems across the Southeastern United States, including:

  • Sipsey River (unregulated)
  • Coosa River (regulated)
  • Tennessee River (large, engineered system)

These sites represent a wide range of morphological and hydrological conditions.


📈 Key Findings

  • DeepLabV3 performed consistently in diverse flows, seasons, and river types.
  • MAT generated smooth, accurate centerlines, even for highly sinuous reaches.
  • The workflow successfully estimated river widths at high spatial consistency.
  • SAR‑based analysis captured changes invisible in optical imagery.
  • The pipeline is fully reproducible and scalable for large river systems.

🧩 Why This Matters

This hybrid approach improves river monitoring by offering:

✔ Reliable measurements under clouds and floods

✔ Automated, repeatable analysis across large regions

✔ Better quantification of geomorphic change

✔ New tools for hydrologists, geomorphologists, and remote sensing scientists

The method can support:

  • Flood early‑warning systems
  • Dam impact assessment
  • River migration prediction models
  • Machine learning–based river classifications
  • Climate‑driven water resource studies

🔬 Conclusion

This research shows that combining deep learning with geometric transforms is a powerful way to map rivers more accurately than ever before.
Using Sentinel‑1A SAR ensures robust monitoring even in challenging conditions.

Our method provides:

  • High‑precision centerlines
  • Reliable width estimates
  • Strong performance across diverse river systems

This approach opens new doors for large‑scale river analysis and future AI‑driven hydrological research.

Links: Earth Surface Processes and Landforms | Geomorphology Journal | Wiley Online Library

Comments