Landscape pattern analysis by applying landscape metrics is a widely used approach to reveal the ecological consequences of landscape change.Soil erosion is the processes of detachment and relocation of soil materials over landscapes,which are regulated by the spatial pattern of landforms,vegetation cover,and human activities.Therefore,landscape pattern analyses to reveal the effects of landscape pattern shifts,especially the land use/cover change,on soil erosion is a good example of the practice of principles and methodology of landscape ecology.However,the capacity of landscape metrics,specifically metrics of "Fragstats" family,used to explain the relationship between the landscape pattern and soil erosion processes was problematic.Therefore,the establishment of principles to select effective metrics is necessary.In this paper,soil erosion and sediment delivery over multiple years in a loess watershed were modeled using WATEM/SEDEM.Representative landscape metrics from four categories,including patch/edge density,shape indices,metrics describing aggregation,and diversity indices were calculated using Fragstat 4.2 based on land use/cover datasets.The correlation between soil erosion and these landscape metrics were tested to evaluate their validity in coupling landscape pattern shifts with soil erosion dynamics.The consistency in revealing the relationship between soil erosion and the spatial pattern of "sink" and "source" landscape types was also evaluated.The results advocate that there are limitations of applying landscape metrics following the patch-corridor-matrix model to indicate the soil erosion status.However,at the landscape class level,landscape metrics showed greater effectiveness in linking landscape pattern with soil erosion.The following principles for selecting suitable landscape metrics for soil erosion status assessments were determined:(1) the signs of correlation coefficients between soil erosion variables and landscape metrics should be different for sink and source types;(2) the signs of correlation coefficients between landscape metrics for multiple sink or source types and soil erosion variables should be the same.Although landscape metrics are relatively effective in coupling landscape pattern with soil erosion at the class level,we show here that using them as predictive tools could result in considerable uncertainty.It was suggested that development of metrics that reflect the behavior of landscape patterns in soil erosion and sediment delivery could reasonably promote the predictive capacity of landscape metrics.