Optimizing dynamic thresholds with salt endmember constraints for lake mapping in arid regions

Arid-region terminal lakes have large saline areas present due to the arid climate and human water extraction. In practical applications, the spectral characteristics of saline areas are easily confused with water bodies, so accurate identification of water body boundaries is of great significance for water resource management, wetland protection, and drought monitoring. In this study, using Ebinur Lake as an example, a novel water index named NDWFIsaline was developed by incorporating a saline endmember into the Normalized Difference Water Fraction Index (NDWFI) framework. A dynamic threshold is set to differentiate between the water body and the saline wetland. Meanwhile, this study utilized normalized difference water index (NDWI), modified normalized difference water index (MNDWI), Enhanced Water Index (EWI), and NDWFI to extract the water bodies. Results indicate that the NDWFIsaline index demonstrated outstanding performance, consistently achieving the highest accuracy (exceeding 97 %) across different seasons, with Kappa coefficients all surpassing 0.94. In contrast, under identical salinity conditions, existing water indices showed slightly inferior and more variable performance: NDWI, EWI, and NDWFI accuracy ranged between 87 % and 98 %, with Kappa values approximately 0.61–0.92; MNDWI accuracy consistently fell below 89 %, and its Kappa coefficient remained below 0.75. Compared to the Global Surface Water (GSW) product, the NDWFIsaline index exhibits a reduction in the leakage rate of approximately 2 % and a significant decrease in the misclassification rate of about 28 %. NDWFIsaline's unique resistance to spectral interference from saline environments enables precise water mapping in arid regions with widespread saline-alkali land.