Abstract: Soil cation exchange capacity (CEC) is the basis for soil fertilization and improvement, and thus an evaluation index of soil quality. Studies on the spatial heterogeneity of soil CEC can provide scientific basis for soil nutrient monitoring, management, and precision agriculture implementation. Spatial distribution of CEC in silt loam soil of Lycium barbarum in Zhongning was investigated based on the analysis of autocorrelation and interaction correlation. The CoCriging, ordinary least squares (OLS), geographically weighted regression Kriging (GWR) and Random Forest (RF) models were applied to perform regression analysis. After that, spatial interpolation mapping effect and the accuracy of the models were compared. The results showed that the average silt loam soil CEC in L. barbarum was 13.12 cmol·kg^-1, belonging to moderately fertility. The distribution of soil CEC had spatial autocorrelation. There were different spatial relationships among CEC, pH, soil organic matter, clay and electrical conductivity in different lag distances. The prediction map of RF model avoided the large fragmentation and obvious mutations which occurred on both sides of the CEC map boundary in the prediction maps of CoKriging, OLS and GWR models. Consequently, soil CEC appeared as a natural and gentle transition in the spatial variation map. The RMSE of the RF model was 33.82%, 20.55% and 19.81% lower than that of CoKriging, OLS and GWR models, and R² was increased by 8.84%, 51.92% and 7.69%, respectively. The RF model considered the location of sample space, and obviouslyimproved the accuracy of interpolation, making the mapping more smooth.

Key words: ecological protection, choice experiment, willingness to pay., random utility theory, contingent valuation method