A New Approach To Predicting Soil Properties Considering The Anisotropy In The Geographical Environment

The spatial distribution information of soil is essential for related research on soil and hydrology.At present,the main approach to obtaining soil distribution information is spatial prediction.The existing spatial prediction methods can be divided into three categories according to their core ideas: The method based on the First Law of Geography,the method based on statistical principle,and the method based on the Third Law of Geography.Methods based on statistical theory and the Third Law of Geography,all need to use soil-environmental conditions to predict soil information,therefore the characterization of environmental conditions becomes particularly important.How to use the soil-environmental information included in the neighborhood to the center pixel is particularly important.Based on the principle that the more similar the geographical environment,the more similar the value of the geographical variables,some scholars have explored a spatial prediction method that considers the similarity of the neighborhood environment.The results show that taking the neighborhood information of the geographical environment into account can improve the accuracy of the prediction.The existing research currently uses a circle-shaped window to calculate the neighborhood information when considering neighborhood information.However,in nature,the influence of geographic environment on soil formation at a certain point in different directions is often different,so that the anisotropy in the geographic environment is ignored.This dissertation proposes a method of predicting soil properties considering the anisotropy of geographic environment.The proposed method calculates the environmental similarity over spatial anisotropic neighborhoods between an unvisited location and each soil sample.The specific research ideas are developed from the following two aspects:(1)determining the size of the anisotropic neighborhood for each environmental covariate at each location: divide the area around each location into eight directions,and determine the size of neighborhood in each direction according to the characteristics of the environment covariate;(2)calculating the similarity of the two points,including the similarity of the two locations and the similarity of the neighborhoods of the two locations.Then,the proposed method estimates the soil property value at each unvisited location to be the average of soil property values of soil samples weighted by the corresponding environmental similarities over neighborhood at the soil sample level.The prediction uncertainty is estimated as well.The proposed method is evaluated through mapping soil organic matter(SOM)content(%)in the topsoil for the Zhuxi River Watershed in Fujian Province.The anisotropy of the geographical environment in different terrain is different.In order to verify the effectiveness of the proposed method,the independent soil sample set is grouped by the aspects of slope,topographical part index(TPI)and relative position index(RPI).The evaluation results show that the proposed method got higher accuracies then the method without considering the anisotropy of the geographical environment,especially at the areas in which the anisotropy of the geographical environment is strong.