Classification of salt marsh vegetation in the Yangtze Estuary using remote sensing and UAV technologies

Abstract

The identification and distribution analysis of salt marsh vegetation in Yangtze Estuary are essential for conservation, management and ecological sustainable development of coastal wetlands in this region. However, the accuracy of identification and analysis depends on the model performance and the number of collected sample points used for training. Therefore, this study develops a Bayesian-optimized random forest (BoRF) classification workflow for salt marsh vegetation mapping and combines it with an unmanned aerial vehicle (UAV)-assisted sample collection strategy. After obtaining the optimal feature combination through the comparative analysis, the vegetation classification and precision evaluation were carried out using multi-temporal Sentinel-2A imagery and the sample points collected by a DJI Mavic 3 M drone, in four representative areas of Yangtze Estuary: Chongming East Beach, Jiu Duan Sha, Nanhui East Beach, and Nanhui Coastal Beach. The experiment results indicate that, among the four compared methods, BoRF achieves the best overall performance, with an overall accuracy of 0.88 and a Kappa coefficient of 0.85, and reduces misclassification among vegetation types with similar spectral characteristics. In addition, the UAV-assisted sample construction strategy improves classification accuracy by approximately 3% compared with conventional field-survey-based sampling. These results demonstrate the practical value of the proposed workflow for salt marsh vegetation classification in complex coastal wetland environments.