Based on the research findings, the current status of micro-scale erosion studies predominantly focuses on exploring the mechanisms of erosion occurrence, while research on specific micro-topographic erosion prediction methods remains insufficient25. This study targets micro-topographies on loess bare slopes and proposes modifications to the Soil Conservation Service Curve Number (SCS-CN) model using measured surface runoff (QT) and surface roughness (SR) to predict runoff. The modified model is then integrated with the Modified Universal Soil Loss Equation (MUSLE) for erosion prediction. The results demonstrate:1)Compared to the original SCS-CN model’s runoff prediction result QO (R2=0.7056), the modified models—SCS-Q (using CN values back-calculated from QT and SCS-SR (modified via SR)—achieved runoff predictions QCN(R2=0.9338) and QSR (R2=0.7691), with accuracy improvements of ?32%? and ?9%?, respectively;2)In micro-topographic erosion prediction, Compared to traditional RUSLE factor combination models, the coupled models showed significant accuracy enhancement. The MUSLE-SR model (coupling SCS-SR with MUSLE) exhibited higher precision (NSE ∈ [0.50, 0.94]) than the MUSLE-Q model (coupling SCS-Q with MUSLE) (NSE ∈ [0.23, 0.94]);3)In micro-topographic erosion prediction, The influence of surface measures on the accuracy of the coupling model (ΔNSE = 63%) is significantly greater than that of rainfall intensity (ΔNSE = 52%) and slope (ΔNSE = 40%); 4)Under micro-topographic conditions, the prediction accuracy of coupled models ?increased significantly with rainfall duration?: accuracy was low in the early stage (before 20 min, R2<0.5) but improved markedly in the later stage (after 20 min, R2>0.8).These findings provide a methodological reference for precise micro-topographic erosion prediction.