Human-related runoff alteration at large reservoir dam sites in China based on counterfactual natural–actual runoff comparisons

Paired Q-CLASS natural and actual runoff reconstructions provide a same-site, same-year reference for human-related departure. Annual series at 907 large-reservoir dam sites in China during 1961–2014 were evaluated using alternative metrics, uncertainty stress tests, multiplicity-controlled temporal tests and cluster diagnostics. The national median relative alteration is \(-0.039\) (site-bootstrap 95% confidence interval: \(-0.044\) to \(-0.035\) ), and 81.59% of sites are negative. Alternative formulations give medians of \(-0.041\) to \(-0.042\) and negative proportions of 82.91–83.35%. Negative medians are strongest in the Haihe, Huaihe and Yellow River basins, while 67.14% of sites are negative at both the \(P_{10}\) and \(P_{90}\) annual-runoff states. After trend-free prewhitening and false-discovery-rate control, 281 sites show decreasing and 80 increasing trends; three significant tests have zero median slope. Five-year block-permutation Pettitt tests retain 246 change points, mainly in the 1980s and 1990s, although the count is block-length sensitive. The leading temporal modes are diffuse, and weak cluster separation (maximum silhouette coefficient 0.162) does not support the former five-type interpretation. Q-CLASS point estimates indicate a prevalent, generally modest negative annual departure, but neither resolve reconstruction uncertainty at individual sites nor identify a single causal process.