[Combining Multi-source Remote Sensing Data and Object-oriented Information Extraction for Arid Wetlands]

Wetlands are the heart of oases in the arid and semi-arid regions of Northwest China, playing a crucial role in climate regulation, water supply, flood storage and drought prevention, biodiversity conservation, and maintaining ecological stability in arid areas. The extraction of wetland information in arid regions provides a rapid and accurate means for monitoring the ecological environment, maintaining biodiversity, and preventing desertification and land degradation. Taking the Ningxia Yinchuan metropolitan area along the Yellow River as the study area, this research uses Sentinel-1 synthetic aperture radar （SAR） imagery, Sentinel-2 optical imagery, and topographic data as data sources. It applies object-oriented wetland information feature extraction methods to explore the importance of red edge, radar, and topographic features in extracting wetlands in arid areas. The feasibility of using the RF-Pearson model to select the optimal combination of features for wetlands in arid regions is verified, combined with the random forest algorithm and BP neural network to extract wetlands in the Ningxia Yinchuan metropolitan area in 2021. The results show that： ① Using the red edge band of Sentinel-2 imagery, the radar beam of Sentinel-1 imagery, and topographic data could effectively promote the identification and acquisition of wetland characteristics in arid regions, improving the overall accuracy of wetlands by 3.27%, 2.14%, and 1.83% compared to spectral indices and geometric features, respectively. ② The classification accuracy of the RF-Pearson model feature selection method was the highest, with the order of importance being： spectral features > geometric features > red edge features > radar features > topographic features. ③ The random forest model （RF） based on feature selection had the best classification effect on wetlands in arid region basins, with an overall accuracy of 89.79% and a Kappa coefficient of 0.842 3, which was higher than that of the BP neural network （BP） classification method, indicating that this method had certain reliability in extracting wetland information in arid regions. ④ Wetlands in the Ningxia Yinchuan metropolitan area mainly included five types： rivers, lakes, tidal flats and marshes, reservoir ponds, and ditches. They were mainly concentrated in Yinchuan City, Pingluo County, Shapotou District, Lingwu City, and Zhongning County. River wetlands dominated the wetlands in arid regions and were a prominent type of wetland in the Ningxia Yinchuan metropolitan area. In the classification results, the area of natural wetlands （rivers, lakes, and tidal flats and marshes） was 1 076.65 km², and the area of artificial wetlands （reservoir ponds, ditches） was 108.18 km², accounting for 90.86% and 9.14% of the total area of the study area, respectively. The research results can provide a scientific basis for monitoring the ecological background environment in arid regions and for ecological protection and high-quality development in the Yellow River Basin.