Research On The Regionalization Of Natural Soil Environment In China Based On Data Driven And Machine Learning

Soil properties and their environmental driving factors show different patterns in different geographical locations,with obvious regional characteristics,which may have an important impact on the migration and transformation of soil organic matter,chemical elements and other substances or the environmental behavior of pollutants.According to the National Tibetan Plateau Scientific Data Center,soil organic matter(0.8-173 g/kg),soil water content(0.05-0.6),porosity(0.4-0.6)and accumulation density(1.11-1.59 kg/m~3)showed significant differentiation across the country.Given these huge differences,it is not appropriate to use common standards for environmental management.In order to realize the scientific and rational regionalization of natural soil environment in China and promote the more precise management of soil resources and environment,a new research model of soil environmental regionalization was established in this study.In this paper,304 articles related to“soil”and“regionalization”were summarized by bibliometrics,and the most important indexes of soil regionalization were determined.Combined with the frequency of the index,AHP is used to establish the weighted index system.Based on the index system and spatial data,three representative clustering algorithms,self-organizing feature mapping(SOFM),fuzzy C-means(FCM)and k-means(KM),were used to cluster the soil data.Four clustering validity indexes(Davis-Bouldin(DB)index,Silhouette index(Sil),Calinski-Harabasz(CH)index and Cluster Quality index(CQI))were used to evaluate the clustering results.The main results are as follows:(1)The bibliometric results show that the most common indexes in soil regionalization research include temperature,precipitation,elevation,soil type,vegetation coverage,cation exchange capacity,etc.Through word frequency statistics and co-occurrence analysis,Finally,14 indexes in 4 categories,including geographical conditions(altitude and slope),climatic conditions(average annual precipitation,average annual temperature,annual accumulated temperature,aridity and evapotranspiration),soil properties(pH value,clay content,organic matter content,bulk density,cation exchange capacity,sand content)and land use(vegetation coverage),were determined as the indexes of natural soil environmental regionalization in China.And use analytic hierarchy process to establish the weighted index system.(2)The results of clustering effectiveness analysis showed that,based on the spatial distribution data of the 14 regional indexes in the whole country,the optimal number of clustering was 13(DB=1.16,Sil=0.78,CH=6.77×10~6)by KM algorithm and FCM algorithm,and the optimal number of clustering was 10(CQI=7.72)by SOFM algorithm.However,SOFM algorithm only identified two soil types in southern China,which is inconsistent with the actual situation that the soil in southern China is mainly divided into primary soil,anthropogenic soil and ferribauxite.Finally,FCM algorithm with cluster number of 13 was selected as the best clustering result.(3)The preliminary clustering results show that the boundaries are largely similar to the famous physical geographic or cultural boundaries in China(such as Tarim Basin,Yunnan-Guizhou Plateau,Northeast Plain and Hu Huanyong Line).Because the boundary division of some regions is too detailed and complicated,based on the principle of maximum similarity within spatial groups and maximum difference between groups,artificial adjustment was made to cluster boundaries,and the natural soil environment in China was finally divided into 12 regions with significant differences.In conclusion,this study used existing high-precision spatial data and unsupervised machine learning methods to define and classify the natural soil environment,and established the regionalization scheme of China’s natural soil environment.The scheme deeply explored the regional characteristics of China’s natural soil environment,and could further serve the regional management of China’s soil resources.At the same time,it is proved that the fuzzy clustering algorithm combined with cluster analysis can reflect the gradual change of soil environment,which provides a set of advancing with The Times,scientific and objective zoning method for the era of big data,and has important reference value for improving soil environmental management according to local conditions in China.This method can also be applied to soil zoning in other countries or regions in the world.