From flood to drought: Integrating water level magnitude and timing to predict floodplain vegetation dynamics in Poyang Lake
Kang Peng, Xianqiang Tang, Rui Li, Danyang Wang, Yanping Hu
Abstract
Hydrological variability is a key driver of floodplain vegetation dynamics, yet current models often overlook the role of event timing. In this study, a temporally explicit two-stage modeling framework was developed by integrating a Gaussian stage–area function with random forest residual correction to separately capture water-level magnitude and sequencing effects. Spatial and temporal cross-validation confirmed the robustness of the approach under varying hydrological regimes, supporting the reliability of subsequent threshold analyses. Vegetation structure was further quantified using landscape metrics under different hydrological states. Vegetation cover peaked at ∼11.2 m stage, while sensitivity analysis revealed that suppression occurred when flood durations ranged from 40 to 100 days or drought-free intervals remained below ∼60 days; recovery was promoted by dry intervals exceeding ∼100 days. Strongly connected sub-lakes exhibited frequent vegetation transitions, whereas isolated or semi-regulated basins maintained greater temporal stability but remained vulnerable to the compound disturbances of 2020–2023, especially the 2022 extreme drought. Incorporating timing-based indicators improved model accuracy (R 2 from 0.33 to 0.81) and provided early-warning signals of resilience erosion. These findings offer a replicable framework for wetland ecosystem monitoring and underscore the importance of disturbance–recovery rhythms in guiding adaptive connectivity management.