Farmland heavy metal pollution source analysis and pollution source influence evaluation method

Abstract

The invention relates to the technical field of heavy metal pollution analysis, in particular to a farmland heavy metal pollution source analysis and pollution source influence evaluation method. The method comprises the following steps: collecting area monitoring data; determining environmental factors of heavy metal enrichment or accumulation in the regional soil sample; carrying out refined analysis on various environmental factors to determine the corresponding environmental factors as environmental driving factors for influencing the heavy metal content change of the regional soil sample; identifying key environment driving factors; determining a non-linear relationship between the heavy metal content of the regional soil sample and the key environment driving factor; and quantifying the critical conditions of the corresponding key environment driving factors, and identifying the soil heavy metal content change under the action of the plurality of key environment driving factors. The spatial distribution condition of soil heavy metal pollution is deeply analyzed, so that the influence of various environmental factors on the soil heavy metal pollution is evaluated. The comprehensive influence condition of more environmental factors on the soil heavy metal can be reasonably reflected.

Description

Technical Field

[0001] The invention belongs to the technical field of heavy metal pollution analysis, and specifically relates to a method for analyzing farmland heavy metal pollution sources and evaluating the impact of pollution sources. Background Art

[0002] Soil heavy metals are important pollutants in agricultural soils due to their toxicity, ubiquity, non-biodegradability, and bioavailability for crop absorption, posing a serious threat to food safety and food security. Soil and permafrost together have the largest terrestrial carbon pool, almost five times the carbon inventory in the atmosphere. Contaminated soil may weaken the carbon cycle function, aggravate climate change and agricultural soil degradation, lead to crop yield reduction, and increase the risk of poverty and malnutrition. Based on the source-sink theory of soil heavy metals, clarifying the relationship between soil heavy metals and environmental factors will help early prevention and control and reduce their impact on socioeconomic and health.

[0003] At present, there are limitations in soil source analysis and source impact research. Although existing studies can identify pollution sources, they have failed to deeply analyze the complex relationship between heavy metals and environmental factors. The complexity of the soil system poses a challenge to exploring the nonlinear relationship between environmental factors and soil heavy metal content. When the change of environmental factors exceeds a certain range or reaches a specific critical value, the heavy metal content in the soil will change dramatically. However, little is known about the nonlinear mechanism by which environmental factors affect the accumulation of heavy metals in soil.

[0004] In addition, soils are constantly changing as environmental factors act through various biophysical and biogeochemical processes at local, regional and global scales. The accumulation of heavy metals in soils is the result of the combined action of multiple factors, among which there are complex dependencies, interactions and feedback loops. The effect of influencing factors on soil heavy metal pollution is not individual but synergistic, and no single factor can independently affect the concentration of heavy metals in soil. Studies have found that when altitude is combined with pH or organic carbon, the effect on cadmium in soil at 194.0m increases compared with the single effect of altitude. However, existing studies only focus on the interaction of two environmental factors, which cannot reasonably reflect the comprehensive impact of more environmental factors on soil heavy metal pollution. Summary of the invention

[0005] In order to solve the above problems, the purpose of the present invention is to provide a method for analyzing the sources of heavy metal pollution in farmland and evaluating the impact of pollution sources. The present invention deeply analyzes the spatial distribution of soil heavy metal pollution to evaluate the impact of multiple environmental factors on soil heavy metal pollution. The present invention can reasonably reflect the comprehensive impact of more environmental factors on soil heavy metals.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows.

[0007] The invention provides a method for analyzing farmland heavy metal pollution sources and evaluating the impact of pollution sources, comprising the following steps: collecting regional monitoring data, wherein the regional monitoring data includes soil heavy metal content and environmental conditions of each soil sampling point; identifying and analyzing the regional monitoring data to determine environmental factors for heavy metal enrichment or accumulation in regional soil samples; respectively performing detailed analysis on various environmental factors to determine corresponding environmental factors as environmental driving factors affecting changes in heavy metal content of regional soil samples; ranking the influence of the determined environmental driving factors on the soil heavy metal content by importance to identify key environmental driving factors; constructing a partial dependence graph to determine the nonlinear relationship between the heavy metal content of regional soil samples and the key environmental driving factors; according to the nonlinear relationship between the heavy metal content of regional soil samples and the key environmental driving factors, quantifying the critical conditions of the corresponding key environmental driving factors, and identifying changes in soil heavy metal content under the action of multiple key environmental driving factors to evaluate the influence of the corresponding key environmental driving factors on soil heavy metals.

[0008] Beneficial effects of the present invention: The present invention collects regional monitoring data, identifies and analyzes the regional monitoring data to determine the environmental factors of heavy metal enrichment or accumulation in regional soil samples; performs detailed analysis on various environmental factors to determine the corresponding environmental factors as environmental driving factors affecting the changes in heavy metal content in regional soil samples; identifies key environmental driving factors by ranking by importance, and determines the nonlinear relationship between the heavy metal content of regional soil samples and key environmental driving factors by constructing a partial dependence graph, and finally quantifies the critical conditions of the corresponding key environmental driving factors, and identifies the changes in soil heavy metal content under the action of multiple key environmental driving factors, thereby constructing a method for analyzing the source of heavy metal pollution in farmland and evaluating the impact of pollution sources. The present invention deeply analyzes the spatial distribution of soil heavy metal pollution to evaluate the impact of multiple environmental factors on soil heavy metals. The present invention can reasonably reflect the comprehensive impact of more environmental factors on soil heavy metals, and solves the problem that existing studies cannot reasonably reflect the comprehensive impact of more environmental factors on soil heavy metal pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a comparison chart of the standardized contribution values ​​of each sampling point factor obtained by running the PMF model. Among them, (a) is the standardized contribution value distribution chart of factor 1; (b) is the standardized contribution value distribution chart of factor 2; (c) is the standardized contribution value distribution chart of factor 3; (d) is the standardized contribution value distribution chart of factor 4; (e) is the standardized contribution value distribution chart of factor 5; (f) is the standardized contribution value distribution chart of factor 6.

[0010] Figure 2It is a scatter plot of the relationship between the measured value and the model prediction value of the random forest model test set. Among them, (a) is a scatter plot of the measured value and the predicted value of Cd; (b) is a scatter plot of the measured value and the predicted value of Cr; (c) is a scatter plot of the measured value and the predicted value of Pb; (d) is a scatter plot of the measured value and the predicted value of Hg; (e) is a scatter plot of the measured value and the predicted value of As; (f) is a scatter plot of the measured value and the predicted value of Cu; (g) is a scatter plot of the measured value and the predicted value of Zn; (h) is a scatter plot of the measured value and the predicted value of Ni. Model represents the model prediction curve; test represents the ideal curve of the model.

[0011] Figure 3 It is a ranking diagram of the relative importance of environmental factors to soil heavy metal concentrations. Among them, (a) to (h) are the ranking diagrams of the relative importance of environmental factors to soil Cd, Cr, As, Cu, Pb, Hg, Zn, and Ni concentrations.

[0012] Figures 4 to 7 It is the nonlinear response diagram of the heavy metal content in cultivated soil to environmental factors. Among them, (a-1)~(a-3) are the nonlinear response diagrams of the Cd content in cultivated soil to environmental factors; (b-1)~(b-3) are the nonlinear response diagrams of the Cr content in cultivated soil to environmental factors; (c-1)~(c-3) are the nonlinear response diagrams of the Pb content in cultivated soil to environmental factors; (d-1)~(d-3) are the nonlinear response diagrams of the Hg content in cultivated soil to environmental factors; (e-1)~(e-3) are the nonlinear response diagrams of the As content in cultivated soil to environmental factors; (f-1)~(f-3) are the nonlinear response diagrams of the Cu content in cultivated soil to environmental factors; (g-1)~(g-3) are the nonlinear response diagrams of the Zn content in cultivated soil to environmental factors; (h-1)~(h-3) are the nonlinear response diagrams of the Ni content in cultivated soil to environmental factors.

[0013] Figures 8 to 11 is the conditional inference tree of each soil heavy metal. Among them, (a) is the conditional inference tree of soil Cd; (b) is the conditional inference tree of soil Cr; (c) is the conditional inference tree of soil Pb; (d) is the conditional inference tree of soil Hg; (e) is the conditional inference tree of soil As; (f) is the conditional inference tree of soil Cu; (g) is the conditional inference tree of soil Zn; (h) is the conditional inference tree of soil Ni. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The methods described in the following embodiments are conventional methods unless otherwise specified; the reagents and materials, unless otherwise specified, can be purchased on the market.

[0015] In order to analyze the spatial distribution of soil heavy metal pollution in more depth, the present invention constructs a method for analyzing the sources of farmland heavy metal pollution and evaluating the impact of pollution sources to evaluate the impact of multiple environmental factors on soil heavy metals.

[0016] The method of the present invention mainly comprises the following steps: collecting regional monitoring data, wherein the regional monitoring data include soil heavy metal content and environmental conditions of each soil sampling point; identifying and analyzing the regional monitoring data to determine environmental factors for heavy metal enrichment or accumulation in regional soil samples; respectively performing detailed analysis on various environmental factors to determine corresponding environmental factors as environmental driving factors affecting changes in heavy metal content in regional soil samples; ranking the influence of the determined environmental driving factors on the soil heavy metal content to identify key environmental driving factors; constructing a partial dependence graph to determine the nonlinear relationship between the heavy metal content of regional soil samples and the key environmental driving factors; according to the nonlinear relationship between the heavy metal content of regional soil samples and the key environmental driving factors, quantifying the critical conditions of the corresponding key environmental driving factors, and identifying changes in soil heavy metal content under the action of multiple key environmental driving factors to evaluate the influence of the corresponding key environmental driving factors on soil heavy metals, thereby constructing a method for analyzing farmland heavy metal pollution sources and evaluating the impact of pollution sources.

[0017] It should be noted that the method of the present invention does not rely on any specific machine learning algorithm or model. The choice of model depends on the specific research problem to be solved.

[0018] The positive definite matrix factorization model, referred to as the PMF model, is an improved factor analysis receptor model developed on the basis of factor analysis. Compared with the traditional factor analysis model, the PMF model imposes non-negative constraints on both factor loadings and factor scores during the solution process to avoid negative values ​​in the results of matrix decomposition, making the obtained source component spectrum and source contribution rate interpretable and with clear physical meaning. In addition, the PMF model does not require the measurement of the source component spectrum, and performs error estimation on each individual data point, making data processing less missed and more accurate and reasonable.

[0019] The random forest model, or RF model for short, is a machine learning algorithm based on decision trees. It uses the Bootstrap resampling method to perform multiple random samplings with replacement from the original data set to generate multiple bootstrap samples of the same size as the original data set. Bootstrap samples are also called Bootstrap samples. Each bootstrap sample can build an independent decision tree. After all decision trees are modeled, when new data is predicted, each decision tree will give a prediction result. Finally, the results of all decision trees are summarized through majority voting to generate the final prediction. This method can reduce the overfitting problem that may exist in a single decision tree and improve the accuracy of the overall prediction.

[0020] In many scientific research fields, machine learning provides researchers with more accurate prediction and estimation capabilities, but its weak interpretability limits in-depth research. Explainable machine learning is based on machine learning models and uses a series of methods to make the results of machine learning models interpretable.

[0021] Partial dependency plots, or PDPs for short, show the marginal effects of one or two features on the predictions of a machine learning model. Partial dependency plots can be used to evaluate how changes in a specific independent variable affect the dependent variable while other independent variables remain unchanged. Specifically, the nonlinear relationship between features and target variables is revealed by calculating the average prediction of the machine learning model for each sample. This visualization method is not only easy to understand, but also has strong explanatory power. It is widely used to show the impact of environmental variables on model predictions and to intuitively display the relationship between targets and features. The application of partial dependency plots improves the interpretability of machine learning models, enabling researchers to better understand the working mechanism and prediction results of the model.

[0022] Conditional Inference Tree, or CIT for short, is a tree-based classification algorithm and a decision tree algorithm for recursive binary splitting. Conditional Inference Tree handles selection bias and overfitting problems by embedding a statistical environment of interchangeable tests to distinguish significant from insignificant improvements. Unlike traditional decision trees, CIT selects variables and split points based on significance tests rather than pure progress or homogeneity measures.

[0023] It should be noted that the method of the present invention is not limited to a specific machine learning algorithm or model, and the above is only an example.

[0024] The evaluation method of the present invention is further described below in conjunction with specific embodiments.

[0025] A method for analyzing the sources of heavy metal pollution in farmland and evaluating the impact of the pollution sources comprises the following steps: Step 1, collect regional monitoring data, the regional monitoring data includes soil heavy metal content and environmental conditions of each soil sampling point. The environmental conditions of each soil sampling point include potential pollution sources of regional soil pollution, and the potential pollution sources of regional soil pollution are pollution sources of potential toxic elements of regional soil pollution. For example, the toxic elements are the heavy metal elements shown in Table 1.

[0026] Step 2: Identify and analyze the regional monitoring data to determine the environmental factors of heavy metal enrichment or accumulation in regional soil samples.

[0027] Step 2.1, a method for identifying and analyzing the regional monitoring data, specifically: The data matrix of heavy metal content in regional soil samples was decomposed into factor score matrix, factor loading matrix and residual matrix using the positive definite matrix factor decomposition model method, and the factor score matrix and factor loading matrix were solved when the objective function was minimized.

[0028] In the PMF model, the data matrix of heavy metal content in regional soil samples is decomposed into factor score matrix, factor loading matrix and residual matrix. The matrix form is: ; Where X is the data matrix of heavy metal content in regional soil samples; G is the factor score matrix; F is the factor loading matrix; E is the residual matrix; n is the number of regional soil samples; m is the number of types of heavy metal elements measured in each regional soil sample; p is the number of main factors, that is, the number of potential pollution sources; G≥0 represents the non-negativity constraint of the factor score matrix; F≥0 represents the non-negativity constraint of the factor loading matrix.

[0029] The calculation formula for decomposing the data matrix of heavy metal content of regional soil samples into factor score matrix, factor loading matrix and residual matrix is: ;in, It is i In the soil samples of the area j The concentration of heavy metal elements; It is k Among the potential sources of pollution j The concentrations of the heavy metal elements are expressed as factor loading matrices; It is k A potential source of pollution i The contribution rate of soil samples in each region is expressed as a factor score matrix; It is i In the soil samples of the area j The residual matrix of the concentration of heavy metal elements.

[0030] The Chinese name of Multilinear Engine 2 is Multilinear Engine 2, Multilinear Engine 2 algorithm, referred to as ME-2 algorithm, is a factor analysis model. The PMF model performs iterative calculations based on the Multilinear Engine 2 algorithm, continuously decomposing the regional soil sample concentration data matrix to obtain the optimal factor score matrix and factor loading matrix. The optimization goal is to minimize the objective function. The calculation formula of the objective function is: ; Where Q is the objective function; For the i Soil samples from the area j The uncertainty of the heavy metal elements; For the i In the soil samples of the area j The residual matrix of the concentration of heavy metal elements.

[0031] Uncertainty can be used to weight each individual data point and must be imported before running the software. The formula for calculating uncertainty is: ; Wherein, U is uncertainty; EF is error fraction; C is the concentration of heavy metal elements in regional soil samples; MDL is detection limit.

[0032] Step 2.2, based on the factor score matrix and factor loading matrix when the objective function is minimized, the environmental factors for the enrichment or accumulation of heavy metals in regional soil samples are determined.

[0033] The embodiment of the present invention characterizes the key pollution sources of heavy metal enrichment or accumulation through environmental factors, so as to determine the direction of heavy metal enrichment or accumulation in regional soil.

[0034] Specifically, the spatial distribution of the factor score matrix of the pollution sources of heavy metal enrichment or accumulation in regional soil samples was characterized by the inverse distance weighted interpolation method, as shown in Tables 1 and 2, to determine the key pollution source types of the corresponding heavy metal enrichment or accumulation, and the key pollution source types are source 1 to source 6.

[0035] The present invention selects Ningbo as the research area, and uses the PMF model to qualitatively analyze the heavy metal pollution sources of farmland soil in the research area, and determine the environmental factors of the enrichment or accumulation of various heavy metal elements. The contribution value of the pollution source factor is quantified by the factor score matrix. The pollution source component spectrum matrix is ​​expressed as a factor loading matrix, and the results are shown in Table 1. The pollution source contribution rate matrix is ​​expressed as a factor score matrix. The results are shown in Table 2.

[0036] Table 1 Pollution source component spectrum matrix of PMF model operation results (mg / kg)

[0037] Table 2 Pollution source contribution rate matrix of PMF model operation results

[0038] Note: For the convenience of description, environmental factor 1 is referred to as factor 1, and the corresponding pollution source 1 is referred to as source 1; environmental factor 2 is referred to as factor 2, and the corresponding pollution source 2 is referred to as source 2; environmental factor 3 is referred to as factor 3, and the corresponding pollution source 3 is referred to as source 3; environmental factor 4 is referred to as factor 4, and the corresponding pollution source 4 is referred to as source 4; environmental factor 5 is referred to as factor 5, and the corresponding pollution source 5 is referred to as source 5; environmental factor 6 is referred to as factor 6, and the corresponding pollution source 6 is referred to as source 6. Each pollution source corresponds to the corresponding environmental factor one by one.

[0039] According to Table 1, Table 2 and Figure 1 , the following characteristic elements about the heavy metal pollution sources of farmland soil in the study area can be obtained: In factor 1, only As has a load, and source 1 is the main source of As. In factor 2, Cd, Hg and Ni all have loads, but the contribution rates of Hg and Ni are both less than 10%, while the contribution rate of Cd is 100%, so source 2 is the main source of Cd. In factor 3, the load proportions of Hg, Cr and Pb are relatively large, and the contribution rate of Pb is 100%, so source 3 is the main source of Hg and Pb. In factor 4, the contribution rate of Cu is 100%, which is significantly higher than Hg and Cr, so source 4 is the main source of Cu. In factor 5, the contribution rates of Cr and Ni are equivalent, so source 5 is the main source of Cr and Ni. In factor 6, the contribution rate of Zn is 100%, which is higher than Ni, so source 6 is the main source of Zn.

[0040] Step 3: Detailed analysis is performed on each type of environmental factor to determine the corresponding environmental factor as the environmental driving factor affecting the change of heavy metal content in regional soil samples.

[0041] After determining the direction of heavy metal enrichment or accumulation in regional soil, different environmental factors are selected for each pollution source for detailed analysis. The corresponding environmental factors are determined based on the results of the detailed analysis, and the determined corresponding environmental factors are used as the natural and human driving forces affecting the changes in the content of heavy metal elements in the soil; among them, the content of heavy metal elements is the response variable, and the natural and human driving forces are the explanatory variables.

[0042] The environmental factors are a collection of soil parent material, soil type, soil pH value, road density, distance to roads, distance to rivers, enterprise density, distance to enterprises, nitrogen fertilizer usage, phosphorus fertilizer usage, potassium fertilizer usage, organic fertilizer usage, pesticide usage and agricultural film usage.

[0043] The load of factor 1 is As. The average soil As value in the study area is lower than the soil background value in Zhejiang Province and the whole country. 99.95% of the sampling points are not polluted by As, indicating that it is less affected by human activities. The distribution map of factor 1 contribution value is similar to the distribution map of soil As content, so factor 1 is a natural source. Soil parent material is the main source of As, but it usually does not exceed 15 mg / kg. However, there are 6 sampling points in the study area with As concentrations exceeding 15 mg / kg, mainly distributed in cultivated land in the eastern to southern coastal areas, which may be related to inappropriate agricultural inputs, such as the use of pesticides and fertilizers.

[0044] The load of factor 2 is Cd, and the excess rate of Cd is 5.46%. The potential ecological risk index shows that most points are at risk of pollution. The coefficient of variation of Cd is 50%, indicating that it is affected by human pollution. The high-value points of the Cd content distribution map and the factor 2 contribution value map are mostly concentrated near the main roads, indicating that Cd pollution is related to traffic emissions. Automobile exhaust accumulates heavy metal Cd in the soil through atmospheric deposition and air dust adsorption; automobile tire wear is also the main source of heavy metal Cd in the soil.

[0045] The loads of factor 3 are mainly Hg and Pb. The correlation coefficient between Hg and Pb is 0.712 at the 99% confidence level, indicating that the two have the same origin. The areas where Hg exceeds the standard are concentrated in the central urban area in the southeast and the northern region, which is related to the distribution of local enterprises. Emissions from metallurgical, chemical and machinery manufacturing enterprises are the main sources of Hg and Pb. Coal combustion and automobile emissions also have a significant impact on the accumulation of Pb and Hg.

[0046] The load of factor 4 is Cu. The Cu pollution in the study area is relatively scattered, but the distribution map of the contribution value of factor 4 shows that the high-value area is relatively continuous. Cu pollution in agricultural land may be related to inappropriate agricultural inputs, especially excessive use of fertilizers. The distribution map of the contribution value of factor 4 shows that there are many high-value points in the northern part of the study area. The waste gas released by mechanical and electronic enterprises in industrial activities such as electroplating also aggravates soil Cu pollution through atmospheric deposition.

[0047] The loads of factor 5 are Cr and Ni, which are extremely significantly positively correlated, indicating that Cr and Ni have a common pollution source. In the single factor evaluation, the pollution of Cr and Ni is slight, mainly affected by the parent material of soil formation. The points where Cr and Ni exceed the standard are concentrated around textile enterprises in the south, and industrial waste gas deposition is the main source of Cr.

[0048] The load of factor 6 is Zn, and the coefficient of variation is 33%, which is the smallest among the heavy metals in the study area, indicating that most areas are not polluted by Zn. Zn is mainly distributed around roads and mechanical and electronic enterprises. Traffic and electroplating plant emissions have an impact on Zn. Therefore, Zn comes from natural background, industrial enterprises and traffic emissions.

[0049] The above analysis results show that the soil heavy metal As in the study area mainly comes from natural background and inappropriate agricultural inputs, such as the use of pesticides and fertilizers; Hg mainly comes from pollutant emissions from industrial enterprises; Cr mainly comes from soil parent materials and emissions from textile enterprises; Cd mainly comes from traffic emissions; Pb mainly comes from emissions from industrial enterprises and transportation; Cu mainly comes from the irrational use of agricultural inputs; Zn comes from natural background, industrial enterprises and transportation emissions; Ni mainly comes from natural background. Therefore, source 1 is a composite source of nature and agriculture; source 2 is a traffic emission source; source 3 is a composite source of industry and traffic emissions; source 4 is a composite source of industry and agriculture; source 5 is a composite source of nature and industry; source 6 is a composite source of nature, industry and traffic emissions.

[0050] Step 4: rank the importance of the impact of the identified environmental driving factors on the soil heavy metal content to identify the key environmental driving factors.

[0051] The method for identifying key environmental driving factors is as follows: constructing a decision tree using the random forest model identification method, using the corresponding out-of-bag data for the decision tree to calculate the first out-of-bag data error; adding noise interference to the environmental factors of all regional soil samples of the out-of-bag data, and calculating the second out-of-bag data error; using the calculation formula of the importance of environmental factors, ranking the importance of the impact of the determined environmental driving factors on the soil heavy metal content, and identifying the key environmental driving factors; the calculation formula of the importance of environmental factors is: ;middle, Indicates the first bag out-of-data error; Indicates the second out-of-bag data error; errr indicates error; OOB indicates out-of-bag data; N tree Represents the number of decision trees.

[0052] Random forest model, referred to as RF model. The randomness of RF model is reflected in the random selection of training set and the random selection of features to be selected. Its construction steps are as follows: (1) The number of samples in the original training set is denoted as K, the original training set is denoted as D, K samples are extracted from the original training set with replacement, and K sub-training sets are constructed, which are denoted as D1, D2, ..., D K , the number of samples in the sub-training set is the same as the number of samples in the original training set; (2) K decision trees are constructed using the K sub-training sets. During the splitting process of the decision tree, certain features are randomly selected from all the candidate features, and then the optimal features are selected for splitting. (3) K decision trees obtain K classification results; (4) The K classification results are voted on to obtain the final classification.

[0053] After each sampling, the samples that were not selected constitute K out-of-bag data, referred to as OOB, which can be used for internal error estimation and as test samples for the RF model. The RF model requires fewer parameters to be adjusted; it is less affected by collinearity and outliers; it is less sensitive to overfitting, missing data, and imbalance problems; and it can generate feature importance rankings and partial dependence graphs to improve the interpretability of the RF model.

[0054] The feature importance ranking involved is used to evaluate the influence of environmental explanatory variables on dependent variables. It is divided into three steps: First, use the corresponding out-of-bag data to calculate the out-of-bag data error of each decision tree in RF, denoted as Second, randomly add noise interference to the features of all samples of out-of-bag data, and calculate its out-of-bag data error again, denoted as Third, the number of decision trees in the random forest model is set to N tree , using the calculation formula of the importance of environmental factors, the impact of the determined environmental driving factors on the soil heavy metal content was ranked in importance, and the key environmental driving factors were identified.

[0055] The specific principle is: if noise is randomly added to a feature, the accuracy of the out-of-bag data will be greatly reduced, which means that this feature has a great influence on the classification result of the sample, that is, its importance is relatively high. The accuracy of the RF model is measured by three indexes: root mean square error, determination coefficient, and mean absolute error; the calculation formulas of the three indexes are as follows: ; Where RMSE is the root mean square error; For the I The true value of the test samples; For the I The predicted value of the test sample; For the I The difference between the true value and the predicted value of the test sample; It is the sum of the squares of the differences between the true value and the predicted value of the M test samples.

[0056] ; ; ; Among them, R 2 is the coefficient of determination, which indicates the correlation between the actual value and the predicted value of the test sample; It is the sum of squares of the difference between the true value and the predicted value of M test samples; is the sum of the squares of the differences between the true value and the mean value of the M test samples; For the I The true value of the test samples; For the IThe predicted value of the test sample; is the average value of the test samples.

[0057] ; Wherein, MAE is the mean absolute error; For the I The true value of the test samples; For the I The predicted value of the test sample; For the I The absolute value of the difference between the true value and the predicted value of the test sample; It is the sum of the absolute values ​​of the differences between the true value and the predicted value of M test samples; ∑ is the summation symbol; M is the number of test samples; I is the number of the test sample; I =1 means the starting number for summation is 1.

[0058] The RF model is implemented using the Random Forest package in R. The Random Forest package is often referred to as the random forest package. The number of decision trees in the RF model is set to 500, the minimum number of test samples for the terminal node is 5, and other parameters are the default values.

[0059] Depend on Figure 2 It can be seen that the R 2 Both the Pearson Index and RMSE are relatively ideal, indicating that it is feasible to use random forest simulation to detect soil heavy metal content.

[0060] Figure 3 The importance of 14 environmental factors to soil heavy metal concentrations is shown in order. The black line represents the most powerful anthropogenic factor. The environmental factors are: parent material; soil type; soil pH; road density, km / km 2 ; Distance from roads, km; Distance from rivers, km; Enterprise density, number / km 2 ; Distance from the enterprise, km 2 ; Nitrogen fertilizer usage, kg / mu; Phosphorus fertilizer usage, kg / mu; Potassium fertilizer usage, kg / mu; Organic fertilizer usage, kg / mu; Pesticide usage, kg / mu; Agricultural film usage, kg / mu.

[0061] The results showed that soil heavy metals were affected by a variety of pollution sources from both natural and human sources. Parent material and soil pH had a great influence on the content of all soil heavy metals. Parent material contributed the most to the content of Cd, Cr, Pb, Cu and Ni, while soil pH contributed the most to the content of As, P<0.001.

[0062] In addition to natural factors, soil heavy metals in the study area are also significantly affected by human factors, mainly from agricultural and industrial activities. Industrial activities have a great impact on all heavy metals except Ni, ranking in the top three. Among them, the biggest factor affecting the concentrations of Hg and Zn is enterprise density. This shows that wastewater, waste gas and solid waste discharged by industrial pollution have a significant impact on the content of heavy metals in the soil of the study area. The amount of fertilizer applied is also considered to be the main factor affecting the concentrations of Hg and Ni. In the study area, non-point source pollution caused by agricultural production processes has begun to affect the enrichment of Hg and Ni in cultivated soil. The amount of phosphate fertilizer used has the greatest impact on Hg concentration, while the amount of nitrogen fertilizer used has the greatest impact on Ni concentration.

[0063] Step 5: construct partial dependence plots to determine the nonlinear relationships between heavy metal contents in regional soil samples and key environmental driving factors.

[0064] Step 5.1, construct a partial dependence plot. A partial dependence plot is referred to as a PDP. After considering the average impact of all other independent variables, the partial dependence plot measures the impact of a specific subset of independent variables on the logarithmic probability or probability of the selected dependent variable, and graphically analyzes the trend and degree of the impact of the independent variable subset on the dependent variable to show the independence between the target and a set of features after excluding all other features. The working principle of the partial dependence plot is as follows: First, 14 environmental factors are selected as input features; two sets are distinguished: the first set, denoted as C, contains all environmental factors; the second set, denoted as S, contains one or two environmental factors that need special attention. Second, a random forest model is trained using a data set containing environmental factors; third, for all features in the first set except those in the second set, as non-target features, the representative values ​​of the non-target features in the training data set are selected and fixed. Fourth, for the environmental factors in the second set, as target features, traverse the possible value range; for each target feature value, keep the values ​​of all other features unchanged, and calculate the predicted value through the machine learning model. Fifth, for each target feature value, all predicted values ​​obtained by combining the same target feature value but different non-target feature values ​​are averaged to obtain the partial dependence value under the target feature value. This process is actually marginalizing or averaging the influence of all features except those in set S. Sixth, different values ​​of the target feature and the corresponding partial dependence values ​​are plotted into a partial dependence graph to show the predicted results of the influence of the feature on the random forest model. Seventh, a key assumption of the partial dependence graph is that the features in the first set are not correlated with the features in the second set. If there is a strong correlation between features, the calculated partial dependence values ​​may contain false or impossible data points, affecting the interpretability of the results.

[0065] The estimation of the partial dependency function can be achieved through the Monte Carlo method, that is, by sampling instances in the training data and calculating the average to approximate. This method increases the robustness of the estimate, but may also introduce a certain amount of randomness. In summary, the partial dependency function calculates the average of the predicted results by fixing the values ​​of all features except the target feature and traversing all possible values ​​of the target feature, thereby revealing the independent influence of the target feature on the model's prediction results. This method is an important tool for understanding and explaining the behavior of machine learning models, especially when dealing with complex and high-dimensional data sets.

[0066] Step 5.2, determine the nonlinear relationship between the heavy metal content of regional soil samples and key environmental driving factors.

[0067] Based on the spatial heterogeneity of soil heavy metal content at the sampling points, the RF model was used to quantitatively identify the key environmental driving factors affecting the heavy metal content in each cultivated land soil. Based on the RF model, a partial dependence diagram was constructed to determine the nonlinear relationship between the heavy metal content in regional soil samples and the key environmental driving factors.

[0068] The embodiment of the present invention mainly determines the response mechanism of regional soil heavy metals to key environmental driving factors through the nonlinear relationship between the heavy metal content of regional soil samples and key environmental driving factors, thereby capturing the detailed process of the response of each heavy metal to a single key environmental driving factor.

[0069] The top three environmental factors with the strongest driving force for soil heavy metals were selected, and partial dependence diagrams were established to analyze the response of soil heavy metals, such as Figures 4 to 7 . In the partial dependence diagram of interaction with parent material, al represents alluvial phase; a2 represents plain alluvial phase; aml represents modern estuarine phase; e1 represents granite alluvium; e10 represents Quaternary Pleistocene red soil; e2 represents rhyolite tuff alluvium; e8 represents residual alluvium of purple sand shale; i represents lagoon phase; m1 represents silt phase; m2 represents silt silt phase; m3 represents silt phase. Parent material has a high influence on all soil heavy metals, but the degree of influence of various types of parent material on individual soil heavy metals is very different. For example, the concentrations of Pb and Zn are higher in alluvial phase soils, while all other soil heavy metals are lower. For Cr, Pb and Hg, their contents are high in silt phase soils, while the opposite is true for Cd, As, Cu, Z and Ni.

[0070] Different soil types significantly affect the Cd content, with red soil, paddy soil and skeleton soil having higher Cd contents. The relationship between heavy metals in each soil and soil pH value is similar, showing an increasing trend with increasing pH value, especially in acidic soils. After the soil pH value exceeds about 8, the heavy metal content begins to decrease with increasing pH value, showing a negative correlation.

[0071] When the enterprise density reaches 2 / km 2 After that, the concentrations of Cr, Pb, Hg, As and Zn did not change much, and their trend changed from the initial rapid rise to a gradual flattening, indicating that the intensity of the impact of enterprise density on the concentration of heavy metals in the soil is not constant. The contents of Cd, Pb and Cu gradually increased with the proximity to the enterprise until they were no longer affected after 18 km² away from the enterprise. The Hg content decreased before the application of phosphate fertilizer reached 10 kg / mu, and then it was positively correlated with the application of phosphate fertilizer. The Ni content increased linearly with the increase in the application of nitrogen fertilizer.

[0072] Step 6: Based on the nonlinear relationship between the heavy metal content of regional soil samples and key environmental driving factors, the critical conditions of the corresponding key environmental driving factors are quantified, and the soil heavy metal content under the action of multiple key environmental driving factors is identified to evaluate the impact of the corresponding key environmental driving factors on soil heavy metals.

[0073] Conditional Inference Tree Model, referred to as CIT model. The working principle of CIT model is: first, calculate the p-value according to the relationship between the output variable and each predictor variable; p is the number of potential pollution sources; second, select the variable with the smallest p-value; third, try all possible binary splits between the dependent variable and the selected variable through permutation test, and select the most significant split; fourth, divide the data set into two groups, and repeat the above steps for each subgroup; fifth, repeat until all splits are insignificant or the minimum node has been reached.

[0074] The embodiment of the present invention mainly quantifies the critical conditions of the corresponding key environmental driving factors through the CIT model, and identifies the soil heavy metal content under the action of multiple key environmental driving factors, so as to evaluate the impact of the corresponding key environmental driving factors on soil heavy metals.

[0075] In the embodiment of the present invention, the method for quantifying the critical conditions of the corresponding key environmental driving factors is specifically: Based on statistical tests, the correlation between each key environmental driving factor and the soil heavy metal content is calculated to obtain the corresponding statistics; the key environmental driving factor with the most significant correlation is selected as the basis for splitting the current node, and the optimal split point of the key environmental driving factor with the most significant correlation is determined as the critical value; according to the critical value, the data set is divided into two subsets, and each subset is further split; the above steps are recursively applied until the correlation of all key environmental driving factors is not significant or the preset stopping condition is reached.

[0076] The results of the conditional inference tree model are as follows Figures 8 to 11As shown, the critical conditions of heavy metal content in soil and key environmental driving factors in the study area are displayed. The splitting conditions are shown on the left and right branches, and the leaf nodes show the number of test samples of different categories, recorded as n, and the average content of heavy metals in the node.

[0077] like Figure 8 For Cd, its CIT has 7 nodes, including 4 terminal nodes. Cd in soil is initially divided into two terminal nodes according to the parent material. The soil Cd content from al, a2, el, el0 and i is higher. When affected by other parent materials except al, a2, el, el0, soil type plays a key role in determining the content of Cd in soil. Red soil and paddy soil show higher Cd concentrations, and this type of soil type is further divided by the distance from the enterprise at node 4. When the distance from the enterprise exceeds 3.08 km, the Cd concentration is higher.

[0078] like Figure 8 According to the CIT model results of soil Cr, it is initially divided into two nodes by the parent material. When the parent material is a2, am2, i, m1, m2 and m3, the enterprise density plays a key role in determining the content of soil Cr. The enterprise density is greater than 0.54 / km 2 When the parent material is a1, e1, e10, e2 and e8, pH value plays a key role. When the soil pH value is not more than 5.06, the Cr content is the lowest.

[0079] like Fig. 9 , the CIT of Pb has 7 nodes, including 4 terminal nodes. The soil Pb is initially divided into two nodes by the parent material. The soil Pb concentration is lowest in the parent materials of a2, aml, e8, ml, m2 and m3. When it is in the parent materials of al, e1, el0 and i, the Pb concentration increases with the increase of enterprise density. When the enterprise density is not higher than 0.13 / km 2 When the distance from the enterprise plays a key role, the Pb concentration is lower at a distance of 5.69 km from the enterprise.

[0080] like Fig. 9 The CIT of Hg shows that enterprise density, soil parent material and phosphate fertilizer use are the key factors affecting Hg content. Enterprise density exceeds 0.56 / km 2 The Hg content in the soil is more than twice that in other areas. 2 Region, parent material and phosphate application have significant effects on soil Hg content. The amount of phosphate application on soils in el, e2 or ml will affect the Hg content in the soil, while in other areas it will hardly be affected by the amount of phosphate application.

[0081] like Fig.10 The CIT of As shows that soil As is affected by soil pH, plant density and parent material. Higher As concentrations occur in cultivated land with soil pH>5.29 and parent material of e2, m2 or m3. Lower soil As concentrations occur in cultivated land with pH≤5.29 and plant density≤0.05 / km 2 in the soil.

[0082] like Fig.10 , Cu's CIT shows that soil Cu is affected by pH, distance from the enterprise, and soil parent material. Higher Cu concentrations were found in cultivated soils with soil parent materials of a2, aml, i, m2, or m3. Lower soil Cu concentrations appeared in soils of al, el, el0, e2, e8, or ml, reaching 23.88 mg / kg at node 6 when pH was ≤5.27 and the distance from the enterprise was greater than 4.63 km.

[0083] like Fig.11 There are three factors that split the CIT of Zn. The enterprise density is the primary factor, reaching 0.56 per km. 2 The Zn content is high in the cultivated soil. When the enterprise density is less than 0.56 / km 2 The soil parent material plays a key role. The cultivated land in areas a2 and e2 showed higher soil Zn content, and the average Zn concentration was higher when the soil pH value was greater than 5.27.

[0084] like Fig.11 , Ni CIT has three segmentation variables: parent material, soil pH, and nitrogen fertilizer application. Node 2 has a higher Ni concentration and appears on cultivated soils of a2, aml, i, m2, or m3, similar to Cu. Node 5 has a lower Ni concentration, provided that it is on cultivated soils of al, el, el0, e2, e8, or ml, with a pH value ≤ 5.30 and a nitrogen fertilizer application of 15.63 kg / mu or less.

[0085] It is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing the sources of heavy metal pollution in farmland and assessing the impact of pollution sources, characterized in that: The following steps are involved: Collecting regional monitoring data, the regional monitoring data including soil heavy metal content and environmental conditions of each soil sampling point; Identify and analyze the regional monitoring data to determine the environmental factors for the enrichment or accumulation of heavy metals in regional soil samples; Each type of environmental factor was analyzed in detail to determine the corresponding environmental factor as the environmental driving factor affecting the change of heavy metal content in regional soil samples; The impact of the identified environmental driving factors on soil heavy metal content was ranked in order of importance to identify the key environmental driving factors; Partial dependence plots were constructed to identify nonlinear relationships between heavy metal content and key environmental drivers in regional soil samples; Based on the nonlinear relationship between the heavy metal content in regional soil samples and key environmental driving factors, the critical conditions of the corresponding key environmental driving factors were quantified, and the changes in soil heavy metal content under the action of multiple key environmental driving factors were identified to evaluate the impact of the corresponding key environmental driving factors on soil heavy metals.

2. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 1, characterized in that: The method for identifying and analyzing the regional monitoring data is specifically: The data matrix of heavy metal content in regional soil samples was decomposed into factor score matrix, factor loading matrix and residual matrix using the positive definite matrix factor decomposition model method, and the factor score matrix and factor loading matrix were solved when the objective function was minimized.

3. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 2, characterized in that: The calculation formula for decomposing the data matrix of heavy metal content of regional soil samples into factor score matrix, factor loading matrix and residual matrix is: ； in, It is i In the soil samples of the area j The concentration of heavy metal elements; It is k Among the potential sources of pollution j The concentrations of the heavy metal elements are expressed as factor loading matrices; It is k A potential source of pollution i The contribution rate of soil samples in each region is expressed as a factor score matrix; It is i In the soil samples of the area j The residual matrix of the concentration of heavy metal elements.

4. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 2, characterized in that: The objective function is calculated as: ； Among them, Q is the objective function; For the i Soil samples from the area j The uncertainty of the heavy metal elements; For the i In the soil samples of the area j The residual matrix of the concentration of heavy metal elements.

5. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 4, characterized in that: The formula for calculating uncertainty is: ； Among them, U is the uncertainty; EF is the error fraction; C is the concentration of heavy metal elements in regional soil samples; MDL is the detection limit.

6. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 1, characterized in that: The environmental factors are a collection of soil parent material, soil type, soil pH value, road density, distance to roads, distance to rivers, enterprise density, distance to enterprises, nitrogen fertilizer usage, phosphorus fertilizer usage, potassium fertilizer usage, organic fertilizer usage, pesticide usage and agricultural film usage.

7. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 6, characterized in that: Methods for identifying key environmental drivers, specifically: A decision tree was constructed using the random forest model recognition method, and the first out-of-bag data error was calculated using the corresponding out-of-bag data for the decision tree; noise interference was added to the environmental factors of soil samples in all regions of the out-of-bag data, and the second out-of-bag data error was calculated; Set the number of decision trees in the random forest model to N tree , using the calculation formula of the importance of environmental factors, the importance of the determined environmental driving factors on the content of heavy metals in soil was ranked, and the key environmental driving factors were identified; The calculation formula for the importance of environmental factors is: ； in, Indicates the first bag out-of-data error; represents the second out-of-bag data error; N tree Represents the number of decision trees.

8. The method for analyzing farmland heavy metal pollution sources and assessing pollution source impacts according to claim 1, characterized in that: The method to quantify the critical conditions of the corresponding key environmental drivers is: Based on statistical tests, the correlation between each key environmental driving factor and soil heavy metal content was calculated to obtain the corresponding statistical quantity; Select the most relevant key environmental driving factor as the basis for splitting the current node, and determine the best split point of the most relevant key environmental driving factor as the critical value; According to the critical value, the data set is divided into two subsets, and each subset is further split; The above steps were applied recursively until the correlations of all key environmental drivers were insignificant or the preset stopping criteria were reached.

Images