Abstract:
To address the accuracy limitations of conventional remote sensing approaches for water body extraction in desert environments,which are commonly affected by mixed pixels and boundary blurring,this study proposes a desert water extraction model that integrates the normalized suspended material index (NSMI)with the multi-index collaborative water extraction method (MICWE),termed MICWE-NSMI.The Ulan Buh Desert was selected as the study area,and Landsat 8 OLI surface reflectance imagery from 2020 was used.Image preprocessing,including cloud removal,radiometric calibration,atmospheric correction,maximum value compositing,and index calculation,was performed on the Google Earth Engine platform.A 30 m buffer zone was constructed along the preliminary water boundaries extracted by MICWE,within which turbid water bodies were further refined by combining inverted NSMI with Otsu adaptive threshold segmentation.A comparison with Sentinel 2 imagery validation results indicates that the proposed MICWE-NSMI model significantly improved the boundary integrity and connectivity of highly turbid waters,narrow tributaries,and flowing water bodies,with an overall accuracy of 93.7%and a Kappa coefficient of 0.82.Compared with conventional index-based methods,the deep learning-based ResUNet model,and the Joint Research Centre global surface water dataset,the proposed model effectively reduces water body fragmentation and misclassification in complex scenarios,including lakes in desert hinterlands,irrigation tributaries,and the Yellow River channel.By integrating conventional water body indices with the spatial heterogeneity of suspended sediment,this study expands the methodological framework for remote sensing identification of highly turbid,fragmented,narrow,and dynamic water bodies,providing a new technical pathway for desert and arid-region water resource monitoring.
Hits54
Downloads16
Related Information