Source-sink Spatial Analysis And Spatio-temporal Simulation Of Soil Heavy Metal Pollution In Regional Scale

Soil is an important hub connecting organic and inorganic components and their interfaces in the ecosystem,which provides water and nutrients for plants and serves as one of the basic components of agricultural production.The soil environment is closely related to the safety of agricultural products.In addition,soil is a non-renewable natural resource;once destroyed,it cannot easily recover over a short period.With the rapid development of the social economy and the continuous advancement of urbanization in China,the impact of human activities on the soil environment is growing day by day,and soil environmental is under considerable threat.In particular,soil heavy metal pollution has become increasingly prominent.In view of soil heavy metal pollution,the Chinese government is gradually implementing related pollution prevention and control work.One of the premises for effective pollution prevention and control is to rapidly identify and manage the spatial and temporal distribution characteristics of soil heavy metals and their pollution sources at the regional scale.This paper takes the south region of the Yangtze River Delta with rapid economic development in recent years as the research area and five soil heavy metal elements（chromium,lead,cadmium,mercury and arsenic）as the research object.We collected multi-temporal sampling data and related environment variables,and used spatial analysis models and machine learning algorithms to carry out the research of soil environmental quality assessment,soil pollution source apportionment and soil pollution dynamic simulation of agricultural land in order to provide some technical support for the prevention and control of soil heavy metal pollution in China.The main research contents and conclusions are as follows:（1）Soil environmental quality assessment usually focuses on a single object（soil,crop or human health）.The assessment results only reflect the pollution of the assessment object,but ignore the interaction and relationship between soil,crops and human health.To address this problem,this study mainly studied soil heavy metal elements,adopted the spatial multi-criteria decision-making model and established a combined evaluation model of soil environmental quality,which mainly focused on human health.This model used the fuzzy method to solve the different dimensions of various indexes,the applied analytic hierarchy process method to define the weight of each index and integrated the results by using the ordered weighted average method.This study found that the pollution degree and main pollution elements of soil environmental quality were significantly different when soil,crop or human health were taken as the assessment objects.The combined evaluation model established in this study focused on human health and also considered soil and crop safety.To some extent,this model achieved the unification of the overall trend of regional soil environmental quality assessment,and the result of the model reflected the assessment results of soil,crop and human health.（2）Traditional methods of soil pollution source apportionment are difficult to accurately identify the pollution sources of different heavy metal elements in soils at regional scale,which are mainly applicable to the management of soil pollution sources,not to the prevention and control of soil pollution.The following 17 environmental variables were considered in this study:soil parent material;soil type;land use;soil p H;soil organic matter;PM_2.5;PM₁₀;population density;application rate of fertilizer（nitrogenous fertilizer,phosphatic fertilizer,potash fertilizer and organic fertilizer）,agricultural film and pesticide;and proximity to polluting enterprises,roads and rivers.And this study used random forest to build the prediction models of the different heavy metals in soils,which well predicted the total concentrations and spatial distributions of the heavy metals.In addition,based on the evaluation results of the importance of variables in the prediction models,the main pollution sources of different heavy metal elements in soils were analyzed quantitatively.Proximity to polluting enterprises and atmospheric sedimentation significantly influenced the concentration distribution of soil heavy metals.The combination of random forest and fuzzy K-means models identified and defined the potential risk areas of soil pollution and allowed for delineation of high-risk,low-risk and safe areas.（3）Enterprise pollution is one of the main causes of soil heavy metal pollution.Due to the alteration of enterprises and little available enterprise information,it is difficult to identify the soil pollution caused by different enterprises at regional scale by traditional methods.This study trained different classification models based on support vector machine,naive Bayesian and artificial neural network algorithms.Multinomial naive Bayesian had the best prediction effect with an accuracy of 86.5%and kappa coefficient of 0.82,which showed that predicted results were basically consistent with the actual results.Freely available geographical data from a search engine（together included more than 190,000 enterprises）were classified into different industry types by Multinomial naive Bayesian.The relationship between the different industry classes and soil cadmium and mercury pollution was explored using bivariate local Moran’s I analysis.The analysis revealed areas in which different industry classes had led to soil pollution.In addition,it was found that the soil cadmium pollution in this study area was mainly caused by the high background value of cadmium in soils.（4）By predicting the heavy metal pollution in the future and changes to its trend,effective measures can be taken to slow down or prevent the occurrence of soil pollution.At present,it is still a lack of effective analysis methods for future soil environmental quality prediction.In this study,convolutional neural networks model and a variety of environmental variables were used to calculate the development probability of different types of soil environmental quality.Then,on the basis of a cellular automata model,combined with development probability,self-adaptive inertia coefficient and roulette wheel selection mechanism,this study constructed a spatio-temporal prediction model of soil environmental quality.The kappa coefficient of the prediction model exceeded0.41,indicating that the results and spatial distribution trend predicted by the model were close to the actual results.Finally,the proportion range of soil pollution area was discussed together with optimistic scenario simulation.The proportion of soil pollution area in 2020,2030,and 2050 were 11.8%–17.3%,6.5%–15.2%and 1.9%–8.3%,respectively,and cadmium was the main pollution element in soils.