Mapping typical salt-marsh species in the Yellow River Delta wetland supported by temporal-spatial-spectral multidimensional features

Abstract

The native salt marsh plants of the Yellow River Delta wetland such as Suaeda salsa and Phragmites australis, providing significant habitats for rare waterfowl, are the key to conserve biodiversity and enhance habitats of this critical wetland. These plants are undergoing severe degradation due to rapid invasion of Spartina alterniflora, which has been a major growing threat to the livelihood of waterfowl and the sustainability of the Yellow River Delta wetland. Monitoring the spatial pattern of salt marsh species is fundamental to the conservation and restoration of the ecological functions in the Yellow River Delta wetland. The development of remote sensing technologies is making a leap forward, particularly the high resolution synthetic aperture radar (SAR), which holds the potential to map heterogeneous wetland regardless of weather. In this study, we developed an innovative framework to map the distribution of salt marsh species with the integration of optical (Sentinel-2) and SAR (Sentinel-1) images. Within this framework, a comprehensive set of features including spectral, spatial and temporal features were considered, and the best feature combination was selected and applied in a random forest classification model to obtain the final map. The results show that the temporal-spectral features combined with the spatial-temporal features of the SAR data can effectively improve the separability of Suaeda salsa, Phragmites australis, and Spartina alterniflora. Compared with using optical or SAR data alone, the combination of optical and SAR data improved the kappa coefficient and the overall classification accuracy by 0.10–0.19 and 6.04–11.61%, respectively. The spatial distribution of the two main native plants and the invasive plant can facilitate ecological restoration of the Yellow River Delta wetland. The framework developed by this study can be efficiently replicated and transferred by similar studies. Our approach lays a solid foundation for intelligent monitoring and management of coastal wetland.