Soil surface roughness (RR) and its spatiotemporal variations are important for understanding soil erosion process, but how the erosion-induced evolution of RR affects flow pathways and sediment connectivity patterns remains unclear. This study used high-resolution topographic data sets from a long-term rainfall simulation experiment in semi-arid environments. A total of approximately 1,500-2,400 mm of rainfall was applied on 2.0 m by 6.1 m plots with stony soil at three slope treatments (5%, 12%, and 20%). The temporal variations in RR, flow pathway length (FL), sinuosity (SS), and longitudinal profile roughness (PR) were investigated. The ratio of RR to slope relief, lambda, was incorporated into a newly proposed sediment connectivity index (RIC). The results showed: (a) RR and PR continuously increased by 73.9% and 34.6% as compared to the initial surfaces, respectively, and steeper slopes developed greater RR and PR; (b) FL and SS increased due to the combined effects of increased RR and surficial rock fragment coverage; (c) RIC decreased because of the increases in FL, RR, and PR; (d) lambda, SS and RIC were found to be appropriate predictors for sediment yield rates with determination coefficients of 0.50, 0.72 and 0.38, respectively, signifying that SS would be a promising parameter for modeling sediment yield. This study also highlighted the important roles of RR and lambda in regulating flow and sediment connectivity, and suggested that stony hillslopes might evolve in a way wherein sediment connectivity and the effects of slope gradient on connectivity will decrease as erosion progresses.