An intelligent method for runoff mutation detection based on adaptive multi-scale transfer entropy

Under global climate change and intensifying human activities, runoff systems exhibit increasingly pronounced abrupt changes, making accurate mutation detection crucial for basin water resources management. Traditional statistical methods such as the Mann-Kendall test rely on univariate analysis and cannot capture nonlinear multi-factor interactions. We propose an intelligent runoff mutation detection method based on Adaptive Multi-Scale Transfer Entropy (AMSTE), which integrates Variational Mode Decomposition (VMD) with transfer entropy theory for the first time to establish a multi-scale causal analysis framework. VMD automatically identifies characteristic timescales of watershed hydrological systems; sliding-window transfer entropy quantifies information transfer intensity from meteorological factors to runoff; Bayesian optimization enables fully automated parameter determination; and an improved spectral clustering algorithm precisely detects mutation points. Using 1961-2020 hydro-meteorological data from the Qiantang River Basin and Yangtze River Basin, we conducted comparative validation across scales. Key findings include: (1) AMSTE successfully identified three major mutation periods (1978-1979, 1997-1998, 2003-2005) in both basins with confidence levels of 0.85-0.96, achieving 78% consistency with Mann-Kendall tests; (2) intra-annual scales dominate in small basins (65.3% contribution), while inter-annual scales increase significantly in large basins (39.5%), revealing clear scale effects; (3) climate change accounts for 65% of variability in the Qiantang River Basin versus 55% from human activities in the Yangtze River Basin, indicating scale-dependent driving mechanisms; (4) post-third-mutation transfer entropy declined by 27.6% (Qiantang) and 13.9% (Yangtze), demonstrating greater resilience in large basins. Sensitivity analysis confirms robust method stability with 77.2% cross-basin parameter transferability. This study provides a novel analytical tool for runoff change detection in complex environments, offering valuable insights for water resources management and climate adaptation strategies.