Classification of Wetland Plant Communities in Poyang Lake Based on Feature Optimization Using Optical and SAR Remote Sensing Imagery

ABSTRACT The plant communities of Poyang Lake, constituting the foundational element of the wetland ecosystem, are integral to crucial ecological processes including energy flow, biodiversity sustenance, water purification and hydrological regulation. Consequently, they serve an irreplaceable function in preserving the stability and ecosystem services of the region. This study uses Landsat 8, Sentinel‐2 optical images and Sentinel‐1 SAR images as data sources to extract spectral reflectance, index features and texture features from optical images as well as radar backscattering features from SAR images, constructing a multidimensional feature dataset. The Recursive Feature Elimination algorithm is employed to perform feature optimization on the dataset. Three classification schemes with different feature combinations are designed, and based on the random forest classifier, the impacts of multisource data fusion and feature optimization on the accuracy of plant community identification are investigated. The results demonstrate that the feature optimization‐based classification scheme attains the highest accuracy, reaching 93.42% overall accuracy with a Kappa coefficient of 0.93. Meanwhile, the optical‐SAR data fusion scheme shows significantly superior performance compared with the optical‐only scheme, delivering a 13.04% enhancement in overall classification accuracy. This study provides a scientific reference for remote sensing classification of wetland plant communities and supports biodiversity conservation and ecological management in the Poyang Lake wetland.