Abstract:  

Obtaining spatial distribution of grassland soil properties accurately and efficiently can provide basic information and reference for grassland resource management.Compared with the traditional soil survey method,the soil environment inference model based on fuzzy logic can improve the efficiency of field sampling and accuracy of predictive mapping,which has been widely used in digital soil mapping.However,due to the soil’s spatial variation and its non-linearity with the environmental conditions,the stability of the existing models are relatively low.The models have not been applied in alpine meadow area.In this study,fuzzy C-means clustering (FCM) was used to predict soil pH in the surface layer of grassland soil within a 4 km² area in Bayanbulak District,Xinjiang Uyghur Autonomous Region,China.Six terrain factors,including elevation,slope,aspect,planform curvature,profile curvature and topographic wetness index,were clustered．Fuzzy membership of 9 groups of environmental factors were derived to position 18 soil samples in the area with membership larger than 0.9.Then pH distribution was predicted with fuzzy membership model.The pH value of study ranged from 7.170 to 8.186 and was consistent with the measured values.The mapping results reflected continuous changing of soil properties with terrain changing.There were 35 individual soil samples (16 equal-interval sampling points,9 cross-sectional sampling points and 10 sampling points according to altitude) collected as validation data set.The agreement coefficients between observed values and predicted values were high,and the accuracy of FCM model is higher than that of Ordinary Kriging method and Inverse Distance Weighted method.FCM and purposive sampling for digital soil mapping is also suitable for small-scale region.This approach is an efficient digital soil mapping method with satisfactory prediction precision using less samples.It could be possibly applied to the areas with the similar landscape conditions．

Key words: font-size:10.5pt, fuzzy clustering')">">fuzzy clustering, purposive sampling, pH, Bayanbulak