An approach for identifying key factors controlling variable source impervious area in heterogeneous urban landscapes under the influence of overland flow path

Urbanization significantly influences urban hydrological responses, yet existing studies focus on identifying and estimating the effective impervious area and often neglecting the interplay of multiple factors influenced by overland flow paths. To address this, the Variable Source Index (VSI) was introduced to assess the variability of impervious areas contributing to runoff of drainage system, and a novel approach was developed to identify the key determinants of VSI. The approach constructed an idealized catchment model to simulate VSIs of potential combinations of factors using Storm Water Management Model (SWMM), optimized Support Vector Regression (SVR) hyperparameters through Bayesian optimization, and interpreted factor importance using SHapley Additive exPlanations (SHAP) values. The results show that the SVR model demonstrated superior predictive accuracy, achieving an R-squared of 0.9390 and a mean squared error of 0.0029, with SHAP values ranking Rainfall (Rain), percentage of non-effective impervious area (Pct), depression storage (D-per) of the receiving pervious area (RPA), and infiltration capacity of RPA (Inf) as the most influential factors on VSI. Rain and Pct were identified as significant enhancers of VSI exhibiting a nonlinear threshold-triggering mechanism. In contrast, Inf and D-per were important in reducing VSI. The impact of various factors on VSI is characterized by a complex set of interrelated and antagonistic dynamics. When applied in a case study, the SVR-predicted VSI closely matched those from SWMM simulations, with a determination coefficient approaching 0.99, validating the efficacy and robustness of SVR. The study provides valuable insights for the enhanced design of urban drainage systems and the improved management of urban stormwater.