In the Anthropocene, large dam construction represents one of the most profound interventions in river morphodynamics, triggering non-linear responses in hydraulic geometry that challenge traditional stationarity-based modeling approaches. This study develops a neural stochastic differential equation (SDE) framework to trace the complete morphodynamic adaptation trajectory of monthly flood season (May–October) reach-scale variables—inflow discharge (Q), reach-averaged channel width (B), depth (H), and slope (J)—in the Jingjiang reach following Three Gorges Dam operation, from initial systemic shock through progressive maturation to a new dynamic equilibrium. The framework employs a log-space ‘physical base + stochastic residual’ paradigm for morphodynamic modeling, integrated with Chronos-T5 large language models for annual driver forecasting and a comprehensive three-tier evaluation system, enabling robust analysis of non-stationary morphodynamic evolution under deep uncertainty. Using a four-stage workflow that calibrates on the post-dam ‘new normal’, retrospectively diagnoses the pre-stabilization transition, validates out-of-sample performance, and explores future scenarios, we reveal quantitative evidence of morphodynamic system maturation. Back-casting over the transitional period (2000–2008) yields coverage as low as 59.3 % (vs. 95 %), quantifying transient dam impacts; comparing a short post-dam training set (2009–2016) with an extended set (2009–2022) shows the annual-driver model shifting from variability-driven indicators to rejecting slow climatic trends as confounding noise. Future projections under Shared Socioeconomic Pathways (SSP245 and SSP585) scenarios unveil divergent morphological adaptation pathways with distinct statistical signatures—SSP245 exhibits sustained incremental adjustment (channel width Mann-Kendall trend strength τ = 0.7298) versus SSP585's shock-adaptation response (τ = 0.7747)—while Dynamic Mode Decomposition indicates identical growth rates for discharge and channel width (0.0014 and 0.0003, respectively) with slight scenario differences for depth and slope, and Convergent Cross Mapping demonstrates enhanced causal coupling under extreme forcing (causal correlation coefficient ρ = 0.986 vs. 0.993). These findings fundamentally advance our understanding of dam-impacted river morphodynamic systems and provide a transferable methodological blueprint for analyzing complex morphodynamic processes undergoing regime transitions in human-dominated landscapes.
扫码关注我们
求助内容:
应助结果提醒方式:
