Quantitative attribution of the drivers of Poyang Lake water level changes based on similarity analysis and the DA–GRU model

As the largest freshwater lake in China, Poyang Lake (PYL) has undergone significant hydrological alterations in recent decades, particularly a continuous decline in autumn water levels, yet the relative contributions of different drivers remain controversial. This study integrates similarity analysis with a Dragonfly Algorithm (DA) optimized Gated Recurrent Unit (GRU) model, forming a control variable framework that explicitly separates timing and magnitude effects of different drivers, enabling quantitative attribution of the effects of the Three Gorges Reservoir (TGR) regulation and channel morphological changes on PYL water level decline.The similarity analysis indicates a structural shift in the hydrological linkage between the Yangtze River and PYL after 2003, marked by a decoupling of mainstream discharge and lake water levels. Scenario simulations indicate that TGR regulation primarily alters the seasonal discharge regime, advancing post-flood water level recession by weakening the backwater effect. In contrast, channel morphological changes, including riverbed incision and cross-sectional enlargement, emerge as the dominant and more persistent control on water level decline. Quantitative attribution shows that about 77% of PYL’s water level decline since 2003 is attributed to channel morphological changes, while about 23% is associated with TGR regulation. Overall, among two primary driving factors, TGR regulation mainly governs the timing of water level decline, while channel morphological changes control its magnitude.