Mapping dominant plant communities in the degraded Zoige swamp using Sentinel-1/2 imagery and its implications for vegetation restoration

Zoige swamp has experienced varying degrees of degradation due to human-driven drainage and global climate change. Through satellite remote sensing, existing studies have obtained data on the distribution and changes in area of degraded Zoige swamp; however, the detailed identification of constituent plant communities has not been conducted. In this study, we used Sentinel-1/2 imagery and digital elevation model (DEM) as the data sources to create median composite images for the fruiting and withering seasons of vegetation. Then, we applied the random forest (RF) algorithm to identify three plant communities dominated by Kobresia, Carex, and Blysmus in the degraded Zoige swamp. The generated map achieved an overall accuracy of 89.2%. Finally, we reconstructed the decision rules based on the trained RF model. The results showed that Carex could be identified through the image characteristics of the fruiting season, whereas the other two communities required dual-phase image features, wherein the identification of Kobresia was particularly challenging. This may be related to the differing water demands of vegetation communities: Kobresia is more drought-tolerant and often mixed with other plateau-based herbaceous plants. This study generated three dominant plant communities in the degraded Zoige swamp, providing new reference methods and data-based support to reveal the current degradation status and its restoration potential.