To better understand how surface roughness affects soil erosion, we investigated the rill distribution of sloping farmland under tillage practices in China's Loess Plateau region. Four tillage treatments (three tillage practices and a non-tillage control) were simulated manually using five slope gradient levels (3˚-20˚) and two rainfall intensities (90 and 120 mm h -1 ). The spatial distribution of rills was extracted and four indices were derived based on Strahler's stream ordering. With increasing slope gradient, random roughness exhibited a decreasing effect on sediment yield, and the interaction effect changed from negative to positive at a critical slope gradient of ∼10˚. Low-order rills gradually merged into high-order rills, forming a complete rill network toward higher slope gradients. The bifurcation ratio of different treatments did not differ considerably. With increasing surface roughness, the length of firstorder rills and their longitudinal river slope decreased. For example, on the 10˚slope under 90 mm h -1 rainfall intensity, the longitudinal river slope of contour plowing, manual hoeing, and contour drilling treatments decreased respectively by 29.4, 38, and 48.9% relative to the value of smooth surface (control) with a rise in rill order. Yet a greater surface roughness had the least effect on the number, length, and longitudinal river slope of third-order rills. A similar result was obtained in that correlation between sediment yield and the length of third-order rills was the weakest overall. The results indicate that surface roughness affects rill distribution and soil erosion on sloping farmland via its effect upon low-orderrill development.
INTRODUCTIONRill erosion is a major soil erosion process caused by water on extensive sloping farmlands and rangelands worldwide, leading to serious soil loss (Kimaro et al., 2008). Rill erosion is a major type of soil erosion, because sediment particles detached from both interrill areas and the rill wetted perimeter are easy to transport in rill channels (Nearing et al., 1997).