Over the period of rainfall, urban green infrastructures (UGI) function like a sponge by absorbing surface runoff as sinks; however, they will shift to sources once their runoff reduction capacities are exceeded. This dynamic of sink-source shifts, and its dependence on the vegetation structure, remain poorly understood, limiting the action of flood-resilient UGI strategies. This study employs MIKE SHE/11 model coupled with statistical analysis for such resolution. Across four scenarios ranging from light to heavy rainfall, we identified regime shifts in UGI system through the decreasing to increasing trends of sink fractions, typically occurring around 13–18 h after rainfall starts. Based on these regime shifts, we categorized the UGI system into vulnerable, reliable, and recoverable components, highlighting its heterogeneous performance. In addition, by examining the influence of vegetation structure on sink–source dynamics, we found that a higher probability of sinks under light rainfalls was associated with a greater leaf area index (LAI) and vegetation height standard deviation (VH<inf>STD</inf>), while green volume (GV) and canopy height (CH) played a more prominent role under heavier rainfalls. Threshold effect analysis further revealed that, a high proportion of the recoverable parts met the thresholds of CH (82 %) and GV (85 %), whereas fewer reached the thresholds of LAI (15 %–19 %) and VH<inf>STD</inf> (3 %–6 %). These findings underscore the importance of enhancing 3D vegetation configuration for UGI to adapt to flood impacts. Our study expects to provide actionable knowledge for understanding, quantification, and management of the runoff sink-source dynamics, informing UGI design and planning to achieve urban flood resilience.