Combined effects of rainfall, structural complexity and human regulation on process connectivity in a plain river network

Yangcheng-Dianmao District, Taihu Lake Basin, China While structural and functional connectivity have been widely studied, process connectivity, a dynamic representation of how flow pathways evolve in response to hydrometeorological and anthropogenic influences, has received far less attention. This oversight limits our understanding of real-time hydrological responses in the different landscapes. We addressed the following questions: What are the spatiotemporal patterns of process connectivity across multiple rainfall events? Can an interpretable machine learning model, based on XGBoost and SHAP, quantify and explain the contributions of hydrological, structural, and anthropogenic factors to these patterns? We proposed a functional data analysis framework to quantify the process connectivity values across 16 river network units in the Yangcheng-Dianmao area of the Taihu Basin. A total of 1052 rainfall-runoff events (2015–2020) were analyzed. Based on this, we used an explainable machine learning model (XGBoost-SHAP) to identify the combined effects of rainfall, structural complexity and human regulation on process connectivity. The structural complexity factors included unit area, fractal dimension, main-channel ratios. High-intensity rainfall events often suppressed connectivity, particularly in areas with dense hydraulic infrastructure. Rainfall amount was identified as the dominant external driver, accounting for 43.1% of the model's explanatory power. Other important factors included unit area (10.5%), main channel area ratio (9.9%), and multifractal indices Δ α (9.5%) and Δ f (8.3%). Human regulation intensity contributed 8.4%, indicating a significant but context-dependent role. Results indicate that process connectivity exhibits combined nonlinear and threshold-like response patterns, reflecting the coupled influences of rainfall forcing, river network structure, and engineering regulation. Specifically, extreme rainfall and high levels of hydraulic regulation were found to suppress connectivity, indicating a transition from rainfall-dominated propagation to regulation-constrained hydrological dynamics in engineered lowland systems. • An explainable XGBoost–SHAP approach is applied to reveal key drivers of process connectivity. • Rainfall is the dominant driver, contributing 43.1% to overall process connectivity. • A threshold-driven and multi-regime response reveals the coupled effects of rainfall, river network structure, and human regulation on process connectivity.