Zaka et al. (2026) Self-supervised learning with multimodal remote sensing data for wetland vegetation mapping in Bosten Lake, China

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2026
• Date: 2026-09-17
• Authors: Muhammad Murtaza Zaka, Alim Samat, Ali Usman, Albert Poponi Maniraho, Arslan Akhtar, Firdavs Vosidov
• DOI: 10.1016/j.rsase.2026.102245

Research Groups

• State Key Laboratory of Ecological Safety and Sustainable Development in Arid Lands, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences (China)
• China-Kazakhstan Joint Laboratory for RS Technology and Application, Al-Farabi Kazakh National University (Kazakhstan)
• Research Centre for Ecology and Environment of Central Asia, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences (China)
• Institute of Water Management, Hydrology and Hydraulic Engineering, University of Natural Resources and Life Sciences (Austria)
• National Engineering Technology Research Center for Desert-Oasis Ecological Construction (China)
• University of Chinese Academy of Sciences (China)

Short Summary

This study proposes a self-supervised learning approach using multimodal remote sensing data to improve vegetation mapping in wetlands with limited labelled samples. The method achieves better results compared to supervised baseline models.

Objective

• Investigate the feasibility of self-supervised learning for vegetation mapping in alpine environments with high spectral similarity among vegetation classes and complex spatial patterns.

Study Configuration

• Spatial Scale: Bosten Lake Alpine Wetland, Xinjiang, China (regional scale)
• Temporal Scale: Seasonal hydrological variation (annual cycle)

Methodology and Data

• Models used: Self-supervised learning methods (Barlow Twins, SimCLR, SwAV, and SimSiam) and ResNet-50 model for pre-training; DeepLabV3+ and HRNet segmentation models
• Data sources: Pléiades, PlanetScope-3, Sentinel-2, Zhuhai-1, and RadarSat-2 SAR imagery

Main Results

• The proposed self-supervised learning approach achieves higher mean Intersection over Union (mIoU) values compared to supervised baseline models.
• Multimodal fusion of optical and SAR data improves segmentation performance.

Contributions

• This study demonstrates the potential of self-supervised representation learning with multi-modal remote sensing for vegetation mapping in wetlands with limited labels, contributing to ecological monitoring and planning in alpine environments.

Funding

• This research was funded by the National Natural Science Foundation of China (Grant No. 42171355) and the Xinjiang Uygur Autonomous Region Science and Technology Plan Project (Grant No. 2021AB01105).

Citation

@article{Zaka2026Selfsupervised,
  author = {Zaka, Muhammad Murtaza and Samat, Alim and Usman, Ali and Maniraho, Albert Poponi and Akhtar, Arslan and Vosidov, Firdavs},
  title = {Self-supervised learning with multimodal remote sensing data for wetland vegetation mapping in Bosten Lake, China},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2026},
  doi = {10.1016/j.rsase.2026.102245},
  url = {https://doi.org/10.1016/j.rsase.2026.102245}
}