| As an important basis for the lives on earth,soil not only supports humans’agricultural production activities,but also serves as major source of nutrients needed by terrestrial organisms.However,with the continuous acceleration of industrialization and urbanization,more and more heavy metal pollutants entered the soil and crops through various ways,such as the deposition of atmospheric soot,sewage irrigation and garbage landfill,which inevitably affected the soil quality and food security,and thus threatened human health.Therefore,it is quite necessary to study the current situation of heavy metals’pollution in soil-plant system,especially for improving soil quality and ensuring food safety.According to the previous the survey of soil pollution,it was found that Cadmium(Cd)was the most polluted heavy metal in China.Thus,the study selected three study areas of different geological background to explore the impact of regional environment factors(such as parent materials,soil types,soil p H,topography,etc.)on Cd enrichment in crop,combining statistical analysis,correlation analysis,machine learning and spatial analysis.And we try to achieve higher accuracy of the quantitative prediction of crop cadmium content.The main research contents and conclusions are as follows:(1)Statistical analysis and correlation analysis of heavy metal content in soil-rice systemThe study carried out statistical analysis for the rice Cd content and different forms of soil Cd content,the results showed that total Cd content in soil of area A,area B and area C were 1.50 mg/kg,0.49 mg/kg,0.68 mg/kg,soil available Cd content were 0.26mg/kg,0.07 mg/kg,0.18 mg/kg,and rice Cd content were 0.83 mg/kg,0.22 mg/kg,0.25mg/kg,and all three study areas existed some extreme high values.In addition,while analyzing the correlativity of heavy metal,it seemed that soil total Cd-soil available Cd,soil total Cd-rice Cd,soil available Cd-rice Cd all had strong correlation,which was consistent with the statistical conclusion of the collected literatures.And the correlation of soil available Cd and rice Cd was stronger than it with soil total Cd.What’s more,the result of partial correlation analysis showed that the difference of Cd content correlation in three study areas was mainly owing to different soil properties.And among the common soil properties,soil p H was proved to be the strongest influence factor in the correlation of Cd content.(2)Analysis of controlling factors of Cd content in riceWith the help of qualitative and quantitative auxiliary variables,a multivariate evaluation index model was constructed,and the results showed good fitting effects.On the basis of the multiple evaluation index system,the study carried out nonlinear principal component analysis on the environmental auxiliary variables with multi-collinearity,forming four principal component factors.Then,the factors were brought into the regression equation one by one,confirming the controlling factors which affected the Cd content in rice among three study areas.The results showed that the key controlling factors of the three regions reflected the regional characteristics to a certain extent.In the screening processes of the main controlling factors,the principal component factors contained several environmental variables with large spatial differentiation in the region were selected as the main controlling factors correspondently,such as the soil type and parent material characteristics of the area A and the regional environmental characteristics of the area B,which clearly reflected the regional characteristics.And it seemed that the distribution of rice Cd content was affected by the environmental factors in different study areas.In view of the interaction among the influencing factors,this study further explored the interaction among the factors,the results showed that the interaction effects including soil p H presented a significant nonlinear enhancement effect,which illustrated the importance of soil p H in soil-rice system.(3)Research on the quantitative prediction model of Cd content in riceThe existing models for estimating Cd content in rice were mostly empirical models and mechanism models,which had some limitations in practical application such as low prediction accuracy and poor universality.The study considered the interaction of factors,spatial non-stationarity and complex nonlinear relationship of data,carried out the neural network model to estimate the rice Cd content and analyzed the differences and the applicability of the models.Spatial regression model was introduced to explore the relationship of rice Cd content and regional environment,exploring the spatial heterogeneity of environmental factors,and the study on spatial mapping of rice Cd was conducted.From the point of nonlinear mapping ability of the models,the research compared three typical models,BPNN,CNN and BP-GA,respectively.The results showed that BP-GA had the most outstanding prediction effect on the rice Cd content,there was a strong correlation between the predicted value and the true value,even the R~2of the validation samples after model training could reach 0.963.In addition,from the perspective of spatial heterogeneity of environmental variables,the GWR model was adopted to process spatial regression analysis and realize the prediction mapping of rice Cd content.It can be concluded that GWR,as a predictive mapping method,had unique advantages in soil attribute analysis and environmental data mapping,providing a new idea for the estimation of rice Cd content.The conclusion of this study can provide an idea of how to delineation the potential area with potential rice pollution risk through soil Cd content and related environmental properties,providing basis for the control of agricultural product safety and quality. |
PDF Full Text Request