Abstract

The wetlands of the Yellow River Delta (YRD) are from a coastal wetland ecosystem that is influenced by both ocean tidal dynamics and the substantial sediment deposits carried by the Yellow River. Therefore, accurately classifying the coastal wetlands in the YRD is essential for their utilization, development and the protection of wetland resources. In this study, by using the sentinel-2 multispectral imagery data, we employ machine learning approach integrated with multi-scale segmentation to improve the accuracy of wetland classification in the YRD. The optimal segmentation scale was determined using the ESP2 tool, with spatial units generated based on a shape and compactness factor. Three machine algorithms—Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM)— and combining with multi-scale segmentation method were evaluated for their classification performance at the spatial-unit level, with an emphasis on analyzing interactions between feature space and segmentation units. The results demonstrate that multi-scale segmentation integrated with machine learning algorithms significantly enhances classification accuracy, confirming its suitability for heterogeneous wetland landscapes classification.