Fusion-Based Water Extraction Using Multiple Spectral Water Indices for Accurate Wetland Mapping

Abstract

Accurate delineation of water surfaces using remote sensing data is essential for water resource monitoring, hydrological analyses, and the assessment of environmental changes. This study proposes a fusion-based approach that combines the strengths of the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (mNDWI), and Automated Water Extraction Index (AWEI), which are widely used for water surface and shoreline delineation from Sentinel-2 imagery. The Alakır Reservoir Lake, located in Antalya Province, Türkiye, was selected as the study area, and Sentinel-2 imagery acquired in 2026 was used. NDWI, mNDWI, and AWEI were first calculated separately, and their outputs were subsequently integrated using a fusion technique to delineate the water surface and shoreline. For accuracy assessment, a reference shoreline was manually delineated from a Sentinel-2 composite image acquired on the same date. A total of 2,048 sample points was generated at 5 m intervals along the reference shoreline, and the shortest distances from these points to the shorelines derived by each method were calculated. The methods were compared in terms of mean positional error, standard deviation, and maximum error. Under the conditions investigated in this study, the fusion approach yielded relatively low positional errors, with a mean error of 7.25 m and a standard deviation of 10.03 m. The corresponding values were 10.36 m and 14.68 m for NDWI, 10.89 m and 13.87 m for mNDWI, and 8.25 m and 11.20 m for AWEI. The results indicate that the proposed fusion approach has the potential to provide relatively accurate and stable shoreline delineation compared with individual index methods, particularly in shoreline transition zones under the investigated conditions.