Wetland mapping of Yellow River Delta wetlands based on multi-feature optimization of Sentinel-1/2 data

The classification of the Yellow River Delta wetlands is crucial for their protection and planning. This study proposes a method for wetland mapping in the Yellow River Delta based on multi-source and multi-feature integration using the GEE platform. First, the Sentinel-1/2 and SRTM data were utilized, combined with the Random Forest Recursive Feature Elimination algorithm (RF_RFE) for feature selection. We compared the classification performance of four machine learning algorithms to determine the best classification scheme, and completed the Yellow River Delta wetland mapping from 2020 to 2023. Research shows that: (1) Multi-source feature collection significantly improves the accuracy of wetland classification, and feature selection can reduce redundant features and further improve classification accuracy; (2) The Random Forest algorithm based on feature selection performs exceptionally well, with an overall classification accuracy of 96.48% and a Kappa coefficient of 0.96, surpassing the classification accuracies of KNN, GBDT and CART; (3) In 2023, the effect of clearing S. alterniflora was significant, and its area was reduced to 9.07 km², which was 48.26 km² less than that in 2022, freeing up living space for native vegetation such as Phragmites. The research results are of great significance for biodiversity conservation and ecological restoration in the Yellow River Delta.