Assessment of the accuracy of automated mapping methods for water surfaces based on spectral indices

Abstract

The paper presents a comprehensive analysis of the effectiveness of Sentinel-2 spectral water indices in combination with automated thresholding using the Otsu method for mapping surface waters in areas affected by man-made disasters. The object of the study was the water area of the former Kakhovka Reservoir, whose landscape after the destruction of the Kak-hovka HPP is characterized by pronounced spectral heterogeneity, the presence of mixed pixels, and a fragmented shoreline. Sentinel-2 optical images of medium spatial resolution were used for the analysis, which made it possible to assess the impact of spatial and structural heterogeneity of landscapes, including fragmented water bodies, wetlands, bare soil, and young vegetation. Four spectral indices – NDWI, MNDWI, AWEIsh, and AWEInsh – were used to evaluate water delineation effectiveness, followed by automatic determination of the optimal threshold using the Otsu method. Comparison of the resulting masks with reference data showed that MNDWI provides the highest accuracy and overall consistency, while the AWEI indices exhibit increased sensi-tivity to small and low-contrast water bodies. NDWI demonstrated limitations under spec-trally complex conditions, particularly in the presence of vegetation and turbid surfaces. The obtained results emphasize the importance of using Sentinel-2 SWIR channels to increase the contrast of water features and confirm the feasibility of combining the MNDWI and AWEI indices. This approach improves the completeness and reliability of mapping in areas affected by man-made disasters. The study demonstrates the potential of automated sat-ellite monitoring methods to support hydro-ecological management in complex anthropogeni-cally modified environments. The proposed methodology can serve as a basis for further work on water mapping, water-use planning, and developing adaptation strategies for water-landscape transformations.