Workflow-dependent variability in reservoir surface-water area estimates from Sentinel-2 and Dynamic World: Yamula Dam Lake, Türkiye

Satellite-derived reservoir area is useful where continuous in situ records are unavailable, but reported extent can depend on source product, compositing, threshold, and post-processing. This study compared five water-mapping rules for Yamula Dam Lake, Türkiye, using 16 spring and late-summer periods from 2017 to 2024. The rules comprised fixed-threshold and Otsu-thresholded Modified Normalized Difference Water Index composites from Sentinel-2 Level-2A reflectance and Dynamic World median water probabilities thresholded at 0.50, 0.60, and 0.70. Outputs were evaluated over a common final valid-pixel domain and subjected to the same reservoir-component filter, although the source products retained distinct observation support. Mapped area ranged from 49.01 to 76.45 km². The mean within-period range was 5.67 km² (9.34%), the mean absolute pairwise difference was 2.83 km², and the mean Jaccard index was 0.943. The intersection of all five rules averaged 56.50 km², whereas the union-minus-intersection envelope averaged 7.45 km². Sensitivity tests showed material threshold dependence and smaller effects from resolution, resampling, and component filtering. Temporal slopes were uncertain and differed by rule and season. The findings support explicit reporting of the complete mapping workflow and sensitivity range.