A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau

The Tibetan Plateau is home to dozens of large braided rivers, a phenomenon that is extremely rare in the global distribution of alluvial rivers. Dynamically monitoring these rivers is crucial for understanding the unique sedimentary processes and hydrodynamic patterns of the plateau. However, the c...

Full description

Saved in:
Bibliographic Details
Main Authors: Xiaolu Liu, Xiaoyi Ma, Shuai Qin, Tang Liu, Chenghu Zhou
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10994330/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849688901481922560
author Xiaolu Liu
Xiaoyi Ma
Shuai Qin
Tang Liu
Chenghu Zhou
author_facet Xiaolu Liu
Xiaoyi Ma
Shuai Qin
Tang Liu
Chenghu Zhou
author_sort Xiaolu Liu
collection DOAJ
description The Tibetan Plateau is home to dozens of large braided rivers, a phenomenon that is extremely rare in the global distribution of alluvial rivers. Dynamically monitoring these rivers is crucial for understanding the unique sedimentary processes and hydrodynamic patterns of the plateau. However, the complex network of water channels and sandbars, coupled with frequent cloud cover, intricate terrain, and numerous branches in the plateau region, poses significant challenges to traditional monitoring and extraction methods. To address these challenges, this study proposes a novel dual-feature encoder deep learning model, DIResU-Net, which integrates Sentinel-1 and Sentinel-2 data to achieve high-precision and long-term extraction of braided rivers. The model employs dual encoders to extract features from optical and radar data, combined with a unified decoder and an attention mechanism for efficient feature fusion. Additionally, a multicomposite loss function was designed to enhance the model's performance. Experimental results demonstrate that the proposed DIResU-Net achieves a high F1-score of 0.87 and IoU of 0.79 under cloud-free conditions, significantly outperforming traditional single-source models. In cloud-covered scenarios, the model maintains robust performance (IoU > 0.73) by leveraging the complementary advantages of Sentinel-1 and Sentinel-2 data. The model also exhibits strong temporal generalization in mapping river morphology from 2019 to 2024, highlighting its value for long-term monitoring and environmental management. Further analysis of morphological parameters—such as river width, channel density, and braiding index—reveals clear seasonal and interannual fluctuations across typical river sections, reflecting the dynamic nature of braided river systems on the plateau. This study provides a scalable framework for high-resolution mapping and long-term monitoring of braided rivers, with implications for hydrological analysis and basin-scale management on the Tibetan Plateau.
format Article
id doaj-art-e2878766d9bb4d96bf1724759705c3d4
institution DOAJ
issn 1939-1404
2151-1535
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
spelling doaj-art-e2878766d9bb4d96bf1724759705c3d42025-08-20T03:21:50ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118127731278510.1109/JSTARS.2025.356827210994330A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan PlateauXiaolu Liu0Xiaoyi Ma1Shuai Qin2Tang Liu3https://orcid.org/0000-0001-7934-7303Chenghu Zhou4https://orcid.org/0000-0003-3331-2302State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, ChinaState Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, ChinaHebei South Canal River Affairs Center, Hebei, ChinaState Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, ChinaState Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, ChinaThe Tibetan Plateau is home to dozens of large braided rivers, a phenomenon that is extremely rare in the global distribution of alluvial rivers. Dynamically monitoring these rivers is crucial for understanding the unique sedimentary processes and hydrodynamic patterns of the plateau. However, the complex network of water channels and sandbars, coupled with frequent cloud cover, intricate terrain, and numerous branches in the plateau region, poses significant challenges to traditional monitoring and extraction methods. To address these challenges, this study proposes a novel dual-feature encoder deep learning model, DIResU-Net, which integrates Sentinel-1 and Sentinel-2 data to achieve high-precision and long-term extraction of braided rivers. The model employs dual encoders to extract features from optical and radar data, combined with a unified decoder and an attention mechanism for efficient feature fusion. Additionally, a multicomposite loss function was designed to enhance the model's performance. Experimental results demonstrate that the proposed DIResU-Net achieves a high F1-score of 0.87 and IoU of 0.79 under cloud-free conditions, significantly outperforming traditional single-source models. In cloud-covered scenarios, the model maintains robust performance (IoU > 0.73) by leveraging the complementary advantages of Sentinel-1 and Sentinel-2 data. The model also exhibits strong temporal generalization in mapping river morphology from 2019 to 2024, highlighting its value for long-term monitoring and environmental management. Further analysis of morphological parameters—such as river width, channel density, and braiding index—reveals clear seasonal and interannual fluctuations across typical river sections, reflecting the dynamic nature of braided river systems on the plateau. This study provides a scalable framework for high-resolution mapping and long-term monitoring of braided rivers, with implications for hydrological analysis and basin-scale management on the Tibetan Plateau.https://ieeexplore.ieee.org/document/10994330/Braided riverscloud-robust classificationdeep learningremote sensing
spellingShingle Xiaolu Liu
Xiaoyi Ma
Shuai Qin
Tang Liu
Chenghu Zhou
A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Braided rivers
cloud-robust classification
deep learning
remote sensing
title A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
title_full A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
title_fullStr A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
title_full_unstemmed A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
title_short A Dual-Source Deep Learning Model for Mapping Wet Anabranches in Braided Rivers on the Tibetan Plateau
title_sort dual source deep learning model for mapping wet anabranches in braided rivers on the tibetan plateau
topic Braided rivers
cloud-robust classification
deep learning
remote sensing
url https://ieeexplore.ieee.org/document/10994330/
work_keys_str_mv AT xiaoluliu adualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT xiaoyima adualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT shuaiqin adualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT tangliu adualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT chenghuzhou adualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT xiaoluliu dualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT xiaoyima dualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT shuaiqin dualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT tangliu dualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau
AT chenghuzhou dualsourcedeeplearningmodelformappingwetanabranchesinbraidedriversonthetibetanplateau