The effects of check dam reservoirs on variations in hydrological regimes commonly result in nonlinear runoff-sediment relationships, which are difficult to describe using current reservoir indicators, particularly for watersheds where floods rise rapidly and huge sediment loads occur. In this study, the evolution of the runoff-sediment relationship was investigated through tests for tendencies and abrupt changes in the Xiliu Valley, a typical hyperconcentrated tributary of the Upper Yellow River on the Northern Loess Plateau, China. Generalized additive models (GAMs) were used to simulate runoff and sediment loads as smooth functions of significant physical covariates including reservoir indices. In comparison with the existing reservoir index (RI) and its additional version (ARI), a sediment-associated reservoir index (SARI) was developed to highlight the advantages of more information on reservoir capacities for both flood control and sediment deposition. The results showed significant downward trends in both annual runoff and sediment series. Alterations in runoff-sediment relationships appeared in approximately 1990, and were mostly dominated by the factors of short-duration storm floods and check dams. GAMs including the SARI exerted more negative effects on sediment yield than on runoff and outperformed the models embracing the RI or ARI. Accordingly, incorporation of the SARI could be advocated under changing environments that are mainly influenced by check dams.