Under climate change, the increasing frequency of dramatic extreme rainfall events is significantly altering hydrological responses in agricultural systems. In contour ridge systems, such extreme rainfall triggers rapid rainwater accumulation in furrows, inducing unique partitioning regime among ponding, infiltration, and runoff. Moreover, the widely used Soil Conservation Service Curve Number (SCS-CN) model lacks the physical adaptability to accurately capture extreme rainfall-runoff process. To investigate the partitioning mechanism of extreme rainfall and enhance runoff modeling, high-intensity simulated rainfall experiments (100 mm h(-1), 70 min duration) were conducted on contour ridging surfaces with distinct microtopography and ridge geometry. Results show that runoff rate increased with the progression of rill and headward erosion, while infiltration and ponding rates declined. The average partitioning ratio of extreme rainfall into infiltration and runoff shifted from 0.87:0.12 during inter-rill erosion to 0.77:0.22 and 0.64:0.36 following headward and rill erosion, respectively. Despite accounting for a minimal portion (<1.5 %) of total rainfall, ponding significantly enhanced infiltration and delayed runoff generation. Ridge width (RW), ridge height (RH), and the interaction between ridge height and row grade (RS) controlled extreme rainfall partitioning, with respective contributions of 22.83 %, 11.52 %, and 24.81 %. The BP-SCS-CN method integrating arithmetic mean of the curve number (CN) adjustment and iterative the initial loss coefficient (lambda) calibration based on RW and RH, improved runoff prediction compared to the original SCS-CN model, but it still inadequately captured runoff generation process. In contrast, the physically modified RWHS-SCS-CN method which incorporated RW, RH, RH*RS, and RW*RH into CN calibration and optimized lambda to 0.05, demonstrated superior predictive performance, achieving the highest Nash efficiency coefficient of 0.64 and the lowest mean relative error of -1.10 %. These findings enhance process-based prediction of extreme rainfall partitioning in contour ridge system, and provide guidance for climate-adaptive agricultural water management strategies.