Digital soil mapping based on the similarity of geographic environment over spatial neighborhoods

Digital soil mapping is an efficient and common way of obtaining soil maps with high accuracy and precision. A representative method is the individual predictive soil mapping (iPSM) method that predicts soil properties by comparing the environmental conditions at the specific location of each individual sample with those at prediction sites. This method has proved to be effective especially with limited samples. However, the iPSM method ignores the impact of the environmental context within a spatial neighborhood on soil properties at the center location. This study proposes an iPSM-neighbor method that considers environmental similarity of spatial neighborhoods to make predictions. Experiments in two study areas show that the proposed method outperformed existing methods (i.e. ordinary kriging, random forest, and iPSM), and reduces the RMSE by up to 33% from the original iPSM method. Evaluation samples of different terrain conditions suggest that the iPSM-neighbor is more effective in mountainous areas. Experiment results attest that the incorporation of environmental similarity over spatial neighborhoods is useful in improving prediction accuracies. Different neighborhood size and annulus width settings provide insights into the impact from characteristics of the neighborhood environment on DSM.