准确辨识模型参数是提升“山区-城区”复合水文系统模拟精度的关键。以福州市江北城区五四片区为研究对象，基于暴雨洪水管理模型（SWMM）对比分析了山区（八一水库）与城区（琴亭湖）汇水区在不同降雨重现期（5~50 a）下的参数敏感性动态演变特征。研究发现：①修正Morris法分析显示，山区洪峰流量主导参数随降雨强度增大发生敏感性转换。低降雨强度（5 a重现期）时受Horton下渗衰减系数（Decay Constant）主导（SN=1.48），高降雨强度（50 a重现期）时转向透水面糙率（N-Perv）主导（SN=0.88）；城区则始终表现出显著的受下垫面影响效应，不透水面糙率及洼地蓄水量调控作用显著。②Sobol敏感度分析表明，在极端降雨情景下，山区参数的一阶敏感度指数与总敏感度指数趋同（差值<0.05），参数独立性效应强；城区参数的总敏感度指数显著高于一阶敏感度（比值达1.5~2.5倍），揭示了强降雨驱动下，地表产汇流与管网水动力过程间存在强烈的非线性耦合交互作用。③山区总径流量对N-Perv呈现极高敏感性，反映了地表阻力调节汇流历时、改变累积下渗量的物理机理；城区因高度硬化下垫面限制，产流响应呈现出多参数联合驱动特征。研究成果揭示了“山区-城区”复合系统产汇流特征的演变逻辑，可为沿海山前城市防洪预警与模型参数率定提供科学依据。

[Objective] The Storm Water Management Model （SWMM） is widely used in urban flood simulation and prediction. Existing studies have primarily focused on drainage systems in plain urban areas，with relatively limited attention to the composite hydrological systems of coastal piedmont cities. This study aims to identify the core driving factors of model responses under different land use types and quantitatively analyze the variation characteristics of dominant parameters with increasing rainfall intensity，providing a scientific basis for constructing high-precision urban flood simulation models and parameter calibration. [Methods] Taking the Jiangbei District of Fuzhou as a case study，we investigated the dynamic evolution of parameter sensitivity within mountainous （Bayi Reservoir） and urban （Qinting Lake） catchments across varying return periods （5-50 a） using the SWMM model. [Results] （1） The modified Morris method revealed a sensitivity transition in the dominant parameters for peak flow in the mountainous catchment with increasing rainfall intensity. The Horton decay constant dominated under low rainfall intensity （5 a，SN=1.48），while the Manning’s roughness of pervious surfaces （N-Perv） dominated under high rainfall intensity （50 a，SN=0.88）. Conversely，urban areas exhibit a consistent land-surface control effect，primarily governed by the Manning’s roughness for impervious areas （N-Imperv） and depression storage. （2） Sobol global sensitivity analysis indicates that under extreme rainfall，the first-order and total sensitivity indices of mountainous parameters are highly convergent （difference<0.05），manifesting strong parameter independence. In urban areas，however，the total sensitivity significantly exceeds the first-order sensitivity （by 1.5 to 2.5 times），revealing intense nonlinear coupling and interactions between surface runoff and hydraulic processes in the drainage network driven by heavy rainfall. 3） The total runoff in mountainous regions shows extreme sensitivity to N-Perv reflecting the physical mechanism where surface resistance regulates runoff travel time and cumulative infiltration. In contrast，due to the constraints of highly impervious surfaces，the urban runoff response is characterized by multi-parameter joint driving. [Conclusions] Rainfall intensity significantly regulates the sensitivity structure of model parameters，and the response intensity of different physical parameters to model output exhibits a nonlinear evolutionary trend with changes in their own values. Extreme rainfall scenarios effectively amplify the dominant role of core driving factors，revealing the mechanism logic of system transition from runoff generation-dominated to runoff concentration-constrained. This finding provides a physical basis for refined calibration of mountain-urban composite hydrological models and spatiotemporally differentiated parameter validation，contributing to improved flood warning accuracy and the scientific design of waterlogging mitigation strategies in mountainous cities under multi-intensity rainfall conditions.