Vegetation and its community structure are among the key factors influencing soil erosion. Near soil surface characteristics (including canopy cover, litter layer, biological soil crusts, and root systems) can influence soil erosion processes to varying degrees, thereby affecting the estimation of the cover and management factor (C). However, the C factor estimated from near soil surface characteristics remains unclear, largely because current studies have neglected the quantitative relationships and impacts of some near soil surface characteristics on soil erosion processes. Therefore, the Poaceae plant of Bothriochloa ischcemum (Linn.). Keng (BI) and the Asteraceae plant of Artemisia vestita Wall. ex Bess (AG) were planted under 5, 10, 15, 20, 25 and 30 plants m(-2) to investigate the effect of aboveground part, biological soil crusts (BSCs) and plant root system on soil erosion. In total, 24 steel tanks were planted, and two tanks were used as bare soil controls. The artificial simulated rainfalls (60 mm h(-1)) were conducted on a 15 degrees slope, and the sediment yield rate were measured. The results showed that planting density and herbaceous species composition exert significant controls on soil erosion processes by regulating near soil surface characteristics. The C factor decreased significantly with increasing planting density for both species, with maximum C values being 2.79 (BI) and 2.91 (AG) times than minimum values. Generally, the C factor of plant root system has bigger value, and it were 2.33 and 8.84 times aboveground part and BSCs for BI, and were 50.56 and 1.78 times aboveground part and BSCs for AG. Correlation analysis indicated that C was significantly negatively correlated with vegetation coverage (VC), aboveground biomass (AGB), BSCs biomass (BSCs-M), and root collar (RA), and power functions were found between C and VC, AGB, BSCs-M, and RA. Structural equation modeling further revealed that planting density strongly promoted root development, while the strongest direct inhibitory effects on C originated from BSCs and root mechanical reinforcement. Based on these relationships, an empirical C model was developed as a power function of aboveground biomass, BSC biomass, and root collar, and the model shows good performance (R-2 = 0.71, p < 0.01).