Liang et al. (2026) WetFramework: A deep learning framework for coastal wetland boundary extraction and inundation frequency estimation

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-03-08
• Authors: Jintao Liang, Yong Zhang, Yi Wang, Chao Chen
• DOI: 10.1016/j.jhydrol.2026.135273

Research Groups

• School of Geophysics and Geomatics, China University of Geosciences, Wuhan, China
• Yangtze River Basin Monitoring Center Station for Soil and Water Conservation, Changjiang Water Resources Commission, Wuhan, Hubei, China
• School of Geography Science and Geomatics Engineering, Suzhou University of Science and Technology, Suzhou, China

Short Summary

This paper introduces WetFramework, a novel deep learning framework that integrates Transformer, Mamba, and wavelet transforms to accurately extract coastal wetland boundaries and quantitatively estimate inundation frequency at microscales, demonstrating superior performance and generalization across diverse coastal regions.

Objective

• To develop a systematic quantitative framework (WetFramework) for mapping microscale dynamics of coastal wetland evolution, specifically focusing on precise boundary delineation and quantitative estimation of inundation rhythms.

Study Configuration

• Spatial Scale: Microscale (pixel-level hydrological responses), tested in four representative coastal regions: Yancheng and Dongying (China), Mont-Saint-Michel Bay (France), and San Francisco Bay (USA).
• Temporal Scale: Intra-annual timescales (high frequency, brief duration inundation events), enabling long-term hydrological modeling and capturing periodic inundation characteristics.

Methodology and Data

• Models used: WetFramework, a deep learning framework comprising:
  • Encoder: Integrated Transformer and Mamba modules with a Token-Driven Attention Mechanism (TDAM).
  • Decoder: Wavelet-Enhanced Reconstruction Module (WERM).
  • Inundation Estimation: Fourier-Based Inundation Estimation Module (FBIEM).
• Data sources: Remote sensing data (implied for wetland identification), tidal-height observations.

Main Results

• WetFramework significantly outperforms state-of-the-art models across multiple evaluation metrics for coastal wetland boundary extraction and inundation frequency estimation.
• The framework exhibits robust cross-regional generalization capabilities, performing well in diverse geographical locations.
• It demonstrates strong dynamic modeling capabilities, providing precise mapping of coastal wetland extents and quantitative expression of inundation rhythms.

Contributions

• Proposes WetFramework, a novel deep learning framework for joint modeling of spatial structure and temporal variation in coastal wetlands.
• Integrates Transformer and Mamba modules with a Token-Driven Attention Mechanism (TDAM) to enhance multiscale feature representation.
• Introduces a Wavelet-Enhanced Reconstruction Module (WERM) to improve spatial structure modeling and optimize boundary delineation.
• Develops a Fourier-Based Inundation Estimation Module (FBIEM) for unsupervised modeling of pixel-level hydrological responses and quantitative expression of inundation rhythms using tidal-height observations.
• Provides an effective paradigm for intelligent remote sensing-based wetland identification and long-term hydrological modeling, supporting inundation-dynamics monitoring and management.

Funding

• Not explicitly stated in the provided text.

Citation

@article{Liang2026WetFramework,
  author = {Liang, Jintao and Zhang, Yong and Wang, Yi and Chen, Chao},
  title = {WetFramework: A deep learning framework for coastal wetland boundary extraction and inundation frequency estimation},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135273},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135273}
}