Abstract:To clarify the non-linear influence mechanism of landscape feature on the sediment yield during the "Grain for Green" project, our study selected 33 hydrological station-controlled sub-basins in the in Hekouzhen Longmen Section of the middle reaches of Yellow River during 2006—2016. We used the boosting regression tree method to analyze the non-linear relationships between landscape drivers and sediment yield modulus of the selected sub-basin. The results showed that the proportion of forest and agricultural land contributed the most to the variation of the sediment yield modulus, which contributing 22.0% and 16.9% respectively. Increasing the proportion of forest land or decreasing the proportion of agricultural land can reduce the sediment yield modulus, but when the proportion of forest land area exceeds 15% or the agricultural land area is less than 20%, the increasing rate of sediment yield modulus becomes stable. Increasing the vegetation coverage of forest land and grass land also has a threshold for mitigating soil erosion, and the threshold value of normalized difference vegetation index (NDVI) in grass land and forest land is 0.5 and 0.3 respectively. When the NDVI value below the threshold, the sediment yield modulus decreases rapidly with increase in NDVI, but the effect of vegetation improvement in grass land and forest land on sediment yield modulus is stable when exceeded the thresholds. Reducing the number of edge density (ED) appropriately is conducive to controlling soil erosion, when the ED maintained at 60/100 hectares can achieve the prevention and control of soil erosion effect. The research reveals the non-linear relationships between landscape characteristics and sediment yield at watershed scale, which can provide an important reference for the reasonable vegetation restoration in the Yellow River.