Clustering-based Study On The Spatial Between Heavy Meatal Pollution In Soil Of Agricultural Products Origin And Enterprises

China has entered the"boom period"of economic growth since 2000,but environmental problems,especially heavy metal pollution in soil,has become increasingly serious when the rapid development of the economy.From the source and pool mechanism of pollution,there are various sources of pollution,especially the pollution emissions of enterprises,but the relationship between pollutions and enterprises is quite complicated.Based on soil pollution samples survey data and enterprises POI data in a specific study area,this study used cluster analysis method to analyze the spatial correlation between heavy metal pollution in soil and enterprises distribution in different industries.At First,the spatial distribution patterns of soil heavy metal pollution and enterprises in different industries were studied.Then,three different clustering methods including Moran index methods,Cubist algorithm and finite mixture model,were used to analyze the relationship between soil heavy metal pollution and enterprises in different regions.The main research contents and conclusions are as follows:(1)Assessment of heavy metal pollution and spatial distribution pattern in soil of agricultural products originThere are 8036 sampling points in the study area,mainly distributed in the north,northeast,south and southwest of the study area.Statistical results show that polluted sites accounted for 35.15%using Nemerow index,and the heavy metal pollution was mainly composed of Cd pollution and Hg pollution.The corresponding polluted sites,respectively,accounted for 17.17%and 13.03%.The pollution areas are mainly located in the surrounding areas of towns and industrial parks and areas with high background content.(2)Study on spatial distribution pattern of enterprisesThis study collected a total of 36,000 enterprise data with 25,000 POI addresses from open channels,including textile industry,Leather,fur and feather products and shoemaking industry,chemical materials and products manufacturing industry,nonferrous metal smelting and calendaring industry,metal products industry,of which textile and metal products industry accounted for 80%.The spatial distribution which inferred by kernel density estimation and natural breaks classification method shows that the textile enterprises are mainly located in the central and northern regions of the study area,while metal products are located in the northeast and southern regions.(3)Cluster analysis based on Moran Index methodBased on the sampling locations of 8036 points and the estimated value of spatial distribution density of enterprises,the experimental results show that there is no correlation between heavy metal pollution index and spatial distribution density of enterprises on the scale of the whole study area,whether or not spatial constraints are introduced.However,according to the LISA clustering results of local Moran index,it is found that there is spatial correlation between heavy metal pollution index and spatial distribution density of enterprises on different sub-regions in study area.(4)Cluster analysis based on Cubist algorithmThe Cubist model is trained based on the basic features of sample points and LISA clustering results of bivariate local Moran methods.The features are selected refer the feature importance metrics,and the rule sets generated by the model is analyzed.The analysis results show that Cubist model can combine the basic information of sample points with LISA clustering results generated by local Moran index,generate rules with practical significance and can choose the feature combinations related to single pollution index.But Cubist algorithm is not a traditional clustering method.It is not only difficult to optimize the number of clusters,but also depends on a target variable to generate rule sets.(5)Cluster analysis based on Finite Mixture ModelBased on the single pollution index and features selected by the Cubist model,the Finite Mixture Model is trained with the number of clusters k is optimized by BIC and the dimension-reduced clustering result visualization.The experimental results show that the region where the pollution source enterprises are located can be concluded according to the geospatial mapping of the clustering results.Compared with the Moran index method,the Finite Mixture Model can take into account the internal relationship between different variables.Compared with the Cubist algorithm,the Finite Mixture Model is more suitable for clustering analysis,although it does not quantify the importance of features.From my perspective,the clustering result with combination of Cubist model and Finite Mixture Model show that the soil heavy metal pollution in some areas is related to the distribution of surrounding enterprises with different industries.This research results can be applied to the identification of pollution sources and monitoring of pollution enterprises,and provide a new perspective for the control of heavy metal pollution in soils of agricultural products origin.