How to obtain more reliable soil-environment knowledge from existing historical samples has become an important scientific issue in digital soil mapping. This article used the method of revising the representativeness of samples to obtain higher accuracy of knowledge. Three different algorithms and the spatial similarity relationship between the covariates of the sample space and the overall spatial environment were used to identify the optimal weights for each sampling point of soil based on the estimation of kernel density. The prediction mapping of the content of organic matter on the surface of soil was used as an example to verify the scientific and validity of the method. The results showed that the revised method reduced RMSE and MAE of multiple linear regression mapping by 10.30% and 12.74%， confirming the feasibility and validity of this method. It will provide technical support for processing the data from sampling points of soil to make full use of historical data and improve the accuracy of mapping soil.