In-situ integration reveals cyanobacteria-specific residence-time signals that Sentinel-2 NDCI/FAI and Landsat NDVI do not resolve in South Korea's Four Rivers

Optical satellite indices are widely used to evaluate impoundment effects on algal blooms, but optical greenness and harmful cyanobacteria are different measurement targets. We test this across the 16 multifunction weirs of South Korea's Four Rivers system, combining Sentinel-2 NDCI and FAI, Landsat NDVI, official harmful-cyanobacteria counts, chlorophyll-a, and residence time τ (144 weir-years, 2017–2025). On a common standardized scale with endpoint-specific weir and year fixed effects, the annual residence-time response is larger for harmful cyanobacteria than for chlorophyll-a by δ = 0.540 (95% CI 0.165 to 0.922; Holm-adjusted p = 0.026); the bloom-season contrast is positive but not formally supported (δ = 0.297). Adjustment for discharge and water temperature does not remove the contrast, which is carried by impounded storage (0.653) rather than discharge (0.191, interval covering zero), so it is not a low-flow artefact. Sentinel-2 NDCI tracks chlorophyll-a closely (Pearson r = 0.684) while its association with harmful cyanobacteria is not separable from zero under weir-level clustering ( r = 0.319); FAI shows the same ordering. That gap is cross-sectional: the between-weir rank difference is 0.656 against −0.018 within weirs, so the indices separate the endpoints between sites, not within a reach over time. A basin-preserving sensitivity is reported under within-basin-year exchangeability, the Landsat NDVI comparison is not robust to the water-mask definition, and two designs could not be estimated. Evaluating harmful-cyanobacteria risk from optical greenness alone can therefore return a false negative when the response is cyanobacteria-specific rather than a general increase in optical biomass.