Hierarchical Environmental Risk Assessment Methods in Complex Multi-Media and Multi-Scale Scenarios

Abstract

This invention discloses a hierarchical environmental risk assessment method for complex multi-media, multi-scale scenarios, belonging to the field of environmental risk assessment technology. The method first obtains the heavy metal concentration in waste residue through leaching experiments, and determines the toxicity level using the single pollution index method and the potential ecological risk index method. Then, it classifies the volume level based on quantile values, classifies the local diffusion risk level using inverse distance weighted interpolation, and classifies the remote receptor sensitivity level based on surrounding land use and water sources. Finally, it integrates multi-source data to construct a spatiotemporal joint analysis model to verify the results. Therefore, this invention, by integrating pollution source toxicity, volume, diffusion risk, and receptor sensitivity for hierarchical risk assessment, proposes for the first time a multi-source data fusion and spatiotemporal joint analysis method, effectively solving the problems of single assessment dimensions, insufficient data fusion, and lack of dynamic verification in existing assessment methods under multi-media, multi-scale scenarios.

Description

Technical Field

[0001] This invention belongs to the field of environmental risk assessment technology, specifically relating to a method for hierarchical environmental risk assessment in complex scenarios involving multiple media and scales. Background Technology

[0002] With the long-term development of mineral resources, the heavy metal pollution caused by the accumulation of historical waste residue has become increasingly prominent, posing a potential threat to the surrounding soil, water bodies, and ecological environment. Currently, there has been some research on environmental risk assessment methods for mine waste residue. For example, the invention patent application CN202210922204.5 proposes a method for assessing the comprehensive environmental pollution risk of abandoned mining areas. This assessment method is based on the analytic hierarchy process (AHP) to classify and score pollution sources and sensitivity indicators in abandoned mining areas. However, its assessment system does not fully consider factors such as the toxicity, volume, and multi-media diffusion of pollution sources, thus failing to reflect the associated risks of pollution migration and receptor exposure in complex scenarios. Furthermore, the invention patent application CN202110962898.0 discloses a method for analyzing pollutant environmental risks based on multi-dimensional evaluation factors. This method optimizes risk factor calculation by introducing an environmental persistence coefficient. However, this method mainly relies on chemical analysis data and lacks the integration of spatial information such as remote sensing images and multi-temporal monitoring, also exhibiting insufficient spatiotemporal dynamics in risk assessment.

[0003] Current technologies still have limitations in comprehensive assessment across multiple media and scale scenarios. For example, the invention patent with application number CN202310489715.7 uses a three-dimensional neural network interpolation model to predict heavy metal concentration distribution. While this can improve the accuracy of spatial analysis, it still does not consider the linkage assessment of pollution source toxicity, local diffusion, and remote receptor sensitivity. The model verification mainly relies on static sampling data, making it difficult to support dynamic risk early warning. Furthermore, traditional methods generally neglect the quantitative impact of pollution source size on risk levels. For instance, the correlation between the scale of waste accumulation and pollutant release potential is not systematically incorporated into the assessment system, leading to inaccurate prioritization of control measures and hindering effective guidance for pollution control and ecological restoration.

[0004] In summary, existing assessment methods suffer from the following main problems: First, the assessment dimensions are limited to single indicators, failing to systematically integrate key levels such as the toxicity intensity of pollution sources, spatial volume distribution, multi-media migration and diffusion patterns, and differences in receptor sensitivity. Second, the fusion efficiency of multi-source heterogeneous data is insufficient, and there are certain technical barriers to the analysis of remote sensing interpretation results, field sampling data, and historical monitoring information. Third, the risk assessment models lack dynamic feedback verification mechanisms, leading to systematic deviations between theoretical predictions and actual pollutant diffusion paths. Therefore, designing a new environmental risk assessment method using multi-dimensional feature coupling and dynamic simulation techniques to accurately identify high-risk areas and formulate scientific control priority allocation strategies has become one of the urgent technical challenges to be addressed. Summary of the Invention

[0005] To address the shortcomings of existing assessment methods in the background art, this invention integrates pollutant toxicity, volume, diffusion risk, and receptor sensitivity, and through multi-source data fusion and model validation, achieves hierarchical risk assessment and precise classification, providing a scientific basis for priority control in environmental management. Simultaneously, by constructing a multi-dimensional assessment indicator system, introducing a spatiotemporal joint analysis model, and establishing comprehensive control rules, it solves the problems of comprehensiveness and accuracy in risk assessment under complex scenarios.

[0006] Based on this, the present invention provides a method for hierarchical assessment of environmental risks in complex multi-media and multi-scale scenarios.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for hierarchical assessment of environmental risks in complex multi-media and multi-scale scenarios, comprising the following steps: Step 1: Toxicity assessment of pollution sources: Heavy metal concentration data in waste residue are obtained through leaching experiments. The individual pollution risk level is calculated using the single pollution index method and the comprehensive ecological risk level is assessed using the potential ecological risk index method. The toxicity level of the pollution source is determined based on the higher value of the two risk levels. Step 2: Pollution source size assessment: Based on the quantile values ​​of pollution source size, the size levels are classified to identify pollution sources that require priority control. Step 3: Local diffusion risk assessment: Spatial interpolation of pollutant concentrations in soil, water bodies, and agricultural products is performed using the inverse distance weighted interpolation method. The local diffusion risk level is then determined by combining the weighted calculation results of the direct diffusion risk index and the indirect diffusion risk index. Step 4: Remote diffusion receptor sensitivity assessment: Analyze the land use types and water source distribution within a 2-kilometer buffer zone around the pollution source to construct a sensitivity index model. Calculate the sensitivity index of all pollution sources using quantile values ​​and classify the remote receptor sensitivity levels. Step 5: Comprehensive Risk Assessment: The toxicity level, volume level, local diffusion risk level, and remote receptor sensitivity level obtained from Steps 1 to 4 are weighted and summed according to preset weights to generate the pollution source priority control level. Step 6: Multi-source data fusion and model validation: Integrate multi-temporal high-resolution remote sensing images, field sampling data, and historical environmental monitoring data to construct a spatiotemporal joint analysis model and validate the reliability of the risk assessment results.

[0008] As a further supplementary explanation of the above technical solution, in step 1, according to PI i The risk level of individual pollution items is divided and mapped to five levels. The calculation formula for the individual pollution index method is as follows: ,in, PI i For a single pollution index, C i For the first i The measured concentrations of several heavy metals, S i For the first i Environmental quality standard limits for each type of heavy metal; In step 1, according to RI The comprehensive ecological risk level is divided into five levels, and the calculation formula for the potential ecological risk index method is as follows: ,in In the formula: For the first i Individual potential risk indices for each heavy metal For the first i The toxicity coefficient of each heavy metal; The higher value between the individual pollution risk level and the comprehensive ecological risk level is used as the toxicity level of the pollution source. ,in, L PI , L RI These are numerical levels of 1-5, representing both individual pollution risk and comprehensive ecological risk, with higher values ​​indicating higher risk.

[0009] As a further supplement to the above technical solution, the specific method for classifying the pollution source volume levels based on the quantile values ​​in step 2 is as follows: Arrange the waste residue volumes of all blocks in ascending order to form a pollution source volume dataset. Calculate the 25%, 50%, 75%, and 90% quantile values ​​of the pollution source volume dataset. Based on the quantile values, classify the pollution source volume into five levels. The quantile values ​​are calculated using the linear interpolation formula: ,in, n For the total number of data, k For quantiles, X This is the sorted dataset.

[0010] As a further supplement to the above technical solution, the calculation formula for the inverse distance weighted interpolation method in step 3 is as follows: Weight ,in, Z ( x 0 () represents the predicted value of pollutant concentration at unknown points. ，Z ( x j Let be the pollutant concentration at the j-th sampling point. d j For unknown points and the first j Euclidean distance of each sampling point ，p Distance attenuation coefficient ， The value ranges from 1 to 3.

[0011] As a further supplement to the above technical solution, in step 3, according to RDI 总 The risk of local spread is divided into five levels, and the formula for calculating the comprehensive spread risk index is as follows: RDI 总 = RDI 直接 0.5+ RDI 间接 0.5, of which, RDI 直接 It serves as a diffusion risk index for soil and water media. RDI 间接 This is a diffusion risk index for agricultural products.

[0012] As a further supplement to the above technical solution, the calculation formula for the sensitivity index model in step 4 is as follows: SI = A 1 ×5+ A 2 ×3+ A 3 ×1+ A 4 × (7+ N ),in SI Indicates the sensitivity index; A 1 、A 2 、A 3 、A 4 These are the area percentages of main agricultural land, other agricultural land, residential land, and water area, respectively, and these percentages are calculated based on the total area of ​​the 2-kilometer buffer zone around the pollution source. N7+N represents the number of water sources within the buffer zone, and 7+N is the water area ratio coefficient. Arrange the sensitivity indices of all pollution sources in ascending order to form a sensitivity index dataset. Calculate the 25%, 50%, 75%, and 90% quantiles of the sensitivity index dataset using the linear interpolation formula in step 2. Divide the sensitivity index into five levels based on the quantiles.

[0013] As a further supplement to the above technical solution, the weight allocation rule in step 5 is as follows: 40% weight for pollution source toxicity level, 20% weight for size level, 25% weight for local diffusion risk level, and 15% weight for remote receptor sensitivity level. The comprehensive risk value is obtained by weighted summation. The priority control level is divided into five levels according to the comprehensive risk value. The comprehensive risk value calculation formula is as follows: CR=W1·G1+W2·G2+W3·G3+W4·G4, where W1, W2, W3, and W4 represent the weights of pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively; G1, G2, G3, and G4 represent the standardized score values ​​of pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively, with a value range of 0-10.

[0014] As a further explanation and limitation of the above technical solution, step 6, multi-source data fusion and model verification, specifically includes the following steps: Step 6.1 Data Acquisition and Preprocessing: Perform geometric correction, radiometric correction, noise removal, band fusion and color mosaic on the remote sensing images to ensure that the spatial resolution is not less than 1 meter; use the Z-score standardization method to eliminate dimensional differences in the sampled data; use linear interpolation to fill missing values ​​in the historical monitoring data and remove invalid data with a time span of more than 10 years. Step 6.2 Feature Extraction and Fusion: Extract normalized vegetation index, normalized water index and texture entropy features from remote sensing images; extract pollutant concentration gradient features and spatial clustering features from sampled data; extract annual average growth rate and seasonal fluctuation features of pollutant concentration from historical data, and construct a multidimensional feature matrix; Step 6.3 Spatiotemporal Joint Analysis Model Construction: Using spectral index, pollutant concentration gradient, and historical trend as input variables, the random forest algorithm is adopted, with the number of decision trees set to 100-500 and the maximum depth to 5-15 layers. Gini impurity is used as the node splitting criterion, and the parameters of splitting feature selection strategy, minimum leaf node sample number, and splitting threshold tolerance are optimized through cross-validation. Model outputs pollution diffusion probability P diffusion With receptor sensitivity score S sensitivity The overall risk weight is calculated using the following formula: ,in, α These are the diffusion probability weighting coefficients. β The coefficients are the sensitivity score weights, and the coefficients satisfy the constraints. α + β =1, α The value range is 0.6-0.8. β The value range is 0.2-0.4; Step 6.4 Model Validation: Calculate the Pearson correlation coefficient, root mean square error, and [other parameters] using cross-validation. Ktuba The coefficients are used to verify the accuracy of the model's predictions and to correct the risk level classification results.

[0015] As a further supplement to the above technical solution, the model verification in step 6.4 adopts the following indicators: The formula for calculating the Pearson correlation coefficient is: , in, x i For the first i The measured concentration values ​​of pollutants at each sampling point y i The model predicts the concentration values ​​for the corresponding points. , These are the average values ​​of the measured value and the predicted value, respectively. n The total number of samples; The formula for calculating the root mean square error is: , in, x i For the first i The measured concentration values ​​of pollutants at each sampling point y i The model predicts the concentration values ​​for the corresponding points; Ktuba The formula for calculating the coefficient is: , in, N The total number of samples, m For the number of risk level categories, O kk For the first k The number of samples correctly classified into risk levels. R k and C k The first two results are the manually interpreted result and the model output result, respectively. k The total number of samples for each risk level Ktuba The closer the coefficient value is to 1, the higher the consistency between the model classification result and the human interpretation result.

[0016] As a further supplement to the above technical solution, the output results of the spatiotemporal joint analysis model in step 6 are used to dynamically update the threshold for classifying pollution source volume levels, optimize the distance attenuation coefficient in the inverse distance weight interpolation method, generate risk heat maps, and identify priority control levels.

[0017] Compared with the prior art, the present invention has the following significant advantages: 1. This invention comprehensively considers key factors such as the toxicity, size, local diffusion risk, and long-range receptor sensitivity of pollutants. By combining the single-pollution index method with the potential ecological risk index method to assess the toxicity of pollutants, it can comprehensively reflect the hazards of single and multiple pollutants. The quantile value classification technology is used to assess the size of pollutants, and in conjunction with other assessment indicators, it achieves risk classification of the entire chain from pollutant source to receptor exposure. At the same time, through scientific and reasonable weight allocation and quantitative assessment, the comprehensiveness and accuracy of the assessment results are further improved.

[0018] 2. This invention proposes for the first time a multi-source data fusion and spatiotemporal joint analysis method, which deeply integrates remote sensing images, sampling data, and historical monitoring data to construct a spatiotemporal joint analysis model. The model is verified and corrected by cross-validation methods to calculate indicators such as Pearson correlation coefficient, root mean square error, and Kappa coefficient, which further enhances the accuracy and reliability of environmental risk assessment and provides solid technical support for dynamic risk early warning in complex scenarios.

[0019] 3. This invention constructs a remote receptor sensitivity assessment model based on land use types and water source distribution within a 2-kilometer buffer zone around the pollution source, accurately identifying potential risks in sensitive areas such as drinking water source protection areas and farmland. Simultaneously, it establishes an objective weight allocation and comprehensive risk calculation mechanism, avoiding the subjectivity of traditional weight setting, providing quantifiable scientific basis for environmental management decisions, and facilitating the formulation of targeted environmental management and protection measures.

[0020] 4. The environmental risk assessment method of this invention is not limited to environmental risk assessment of mining waste. Its multi-dimensional assessment approach, data fusion method, and model construction method are universal. It can be widely applied to other fields such as environmental pollution risk assessment of chemical industrial parks and environmental risk assessment around urban landfills, opening up new avenues and providing new technologies and research ideas for risk assessment in different scenarios. Attached Figure Description

[0021] Picture 1 This is a flowchart of the step-by-step environmental risk assessment method in this invention.

[0022] Picture 2 This is a flowchart of the pollution local diffusion risk assessment method in an embodiment of the present invention.

[0023] Picture 3This is a schematic diagram illustrating the risk level of localized pollution diffusion in an embodiment of the present invention.

[0024] Picture 4 This is a schematic diagram of the environmental risk heat in an embodiment of the present invention. Detailed Implementation

[0025] To further illustrate the technical solution of the present invention, the present invention will be further described below through embodiments.

[0026] A hierarchical environmental risk assessment method for complex multi-media and multi-scale scenarios is proposed. It mainly achieves risk classification of the entire chain from pollution source to receptor by coupling assessment of local diffusion and remote receptor, dynamic fusion of multi-source data and adaptive adjustment of weights.

[0027] As attached Picture 1 As shown, the evaluation method includes the following steps: Step 1: Toxicity assessment of pollution sources: Heavy metal concentration data in waste residue are obtained through leaching experiments. The individual pollution risk level is calculated using the single pollution index method and the comprehensive ecological risk level is assessed using the potential ecological risk index method. The toxicity level of the pollution source is determined based on the higher value of the two risk levels. Step 2: Pollution source size assessment: Based on the quantile values ​​of pollution source size, the size levels are classified to identify pollution sources that require priority control. Step 3: Local diffusion risk assessment: Spatial interpolation of pollutant concentrations in soil, water bodies, and agricultural products is performed using the inverse distance weighted interpolation method. The local diffusion risk level is then determined by combining the weighted calculation results of the direct diffusion risk index and the indirect diffusion risk index. Step 4: Remote diffusion receptor sensitivity assessment: Analyze the land use types and water source distribution within a 2-kilometer buffer zone around the pollution source to construct a sensitivity index model. Calculate the sensitivity index of all pollution sources using quantile values ​​and classify the remote receptor sensitivity levels. Step 5: Comprehensive Risk Assessment: The toxicity level, volume level, local diffusion risk level, and remote receptor sensitivity level obtained from Steps 1 to 4 are weighted and summed according to preset weights to generate the pollution source priority control level. Step 6: Multi-source data fusion and model validation: Integrate multi-temporal high-resolution remote sensing images, field sampling data, and historical environmental monitoring data to construct a spatiotemporal joint analysis model. The output of the spatiotemporal joint analysis model is used to dynamically update the threshold for classifying pollution source volume levels, optimize the distance attenuation coefficient in the inverse distance weighted interpolation method, generate risk heat maps, and identify priority control levels, thereby validating the reliability of the risk assessment results.

[0028] In this embodiment, the calculation formula for the single pollution index method is as follows: ,in, PI iFor a single pollution index, C i For the first i The measured concentrations of several heavy metals, S i For the first i The environmental quality standard limits for each heavy metal are determined with reference to the screening values ​​for the corresponding construction land type in GB36600-2018 "Soil Environmental Quality Standard for Construction Land Soil Pollution Risk Control".

[0029] according to PI i Individual pollution risk levels are classified and mapped to the following five levels: safe level, alert level, light pollution, moderate pollution, and heavy pollution.

[0030] Table 1. Standards for Classifying Heavy Metal Pollution Risk Levels Using the Single Pollution Index Method

[0031] In step 1, according to RI The comprehensive ecological risk level is divided into five levels, and the calculation formula for the potential ecological risk index method is as follows: ,in In the formula: For the first i Individual potential risk indices for each heavy metal For the first i The toxicity coefficients of the heavy metals were determined with reference to GB36600-2018 "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard" and toxicological studies, and the values ​​were: arsenic 10, cadmium 30, hexavalent chromium 2, copper 5, lead 5, mercury 40, and nickel 5.

[0032] according to RI The comprehensive ecological risk level is divided and mapped to the following five levels: minor risk, moderate risk, high risk, relatively high risk, and extremely high risk.

[0033] Table 2. Criteria for Classifying Heavy Metal Pollution Risk Levels Using the Potential Ecological Risk Index Assessment Method

[0034] The higher value between the individual pollution risk level and the comprehensive ecological risk level is used as the toxicity level of the pollution source. ,in, L PI , L RI These are numerical levels of 1-5, representing both individual pollution risk and comprehensive ecological risk, with higher values ​​indicating higher risk.

[0035] The specific toxicity levels and numerical ranges of the pollution sources are as follows: Security level: The numerical range is PI ≤0.7 and RI <150 corresponds to a single pollution risk level and a comprehensive ecological risk level of level 1, indicating the lowest risk.

[0036] Alert level: Value range is 0.7 < PI ≤1.0 or 150≤ RI If the value is less than 300 and the larger of the two values ​​is at level 2, it indicates a potential pollution risk and requires enhanced monitoring.

[0037] Light pollution level: numerical range is 1.0 < PI ≤2.0 or 300≤ RI If the value is less than 600 and the larger of the two values ​​is at level 3, there is a certain pollution impact, and preliminary control measures need to be taken.

[0038] Moderate pollution level: numerical value range is 2.0 < PI ≤3.0 or 600≤ RI If the value is less than 800 and the larger of the two values ​​is at level 4, the pollution risk is high and targeted treatment is required.

[0039] Severe pollution level: numerical range is PI >3.0 or RI If the value is ≥800 and the larger of the two values ​​is at level 5, the pollution risk is extremely high and emergency control should be prioritized.

[0040] In this embodiment, waste residue accumulation occupies a large area of ​​land, affecting not only the utilization of land resources but also becoming a source of environmental pollution. The larger the volume of waste residue, the greater its harm to the environment and human health. Therefore, it is necessary to comprehensively consider multiple factors such as pollutant toxicity, pollutant leaching amount, and waste residue volume to more accurately assess the potential ecological risks of waste residue pollution sources. The specific method for classifying the volume levels based on quantile values ​​in step 2 is as follows: Arrange the waste residue volumes of all blocks in ascending order to form a pollution source volume dataset. Calculate the 25%, 50%, 75%, and 90% quantile values ​​of the pollution source volume dataset. Based on the quantile values, classify the pollution source volume into five levels. The quantile values ​​are calculated using the linear interpolation formula: ,in, n For the total number of data, k For quantiles, X Given the sorted dataset; calculate the 25%, 50%, 75%, and 90% quantiles of the pollution source volume dataset, thus obtaining... P 25 , P 50 , P 75 , P90 As a threshold, based on quantile values P k Pollution sources are classified into the following five levels based on their size: extremely small size V < P 25 Smaller size P 25 ≤ V <P 50 Medium size, P 50 ≤ V <P 75 Larger volume P 75 ≤ V <P 90 Extremely large V ≥P 90 This clarifies which areas require priority management due to their large volume of waste residue.

[0041] As attached Picture 2 As shown, the migration and diffusion of heavy metal pollution in waste residue is categorized into two types based on media type: the first is migration and diffusion in natural environmental media such as soil, surface water, groundwater, and sediment; the second is migration and diffusion from natural environmental media to agricultural products related to human health. Five typical heavy metal pollutants—mercury, cadmium, arsenic, lead, and chromium—are selected as indicators for regional heavy metal pollution diffusion risk assessment. For direct pollution, detection data from four natural environmental media (soil, surface water, groundwater, and sediment) are used to calculate the direct pollution level. Since different samples have different units, the single-item pollution index method is used to calculate the pollution level for each type of sample to eliminate the influence of dimensional differences, thus forming a unified dimensionless pollution level data. Single-item pollution level calculations are performed for mercury, cadmium, arsenic, lead, chromium, and characteristic pollutants in natural environmental media. For indirect pollution, detection data from agricultural product samples are used to calculate the indirect pollution level, and the single-item pollution index method is used to calculate the pollution level of agricultural products. Direct pollution levels for the five heavy metals in agricultural products—mercury, cadmium, arsenic, lead, and chromium—are calculated separately. Based on the characteristic that the diffusion of heavy metal pollution decreases with distance, the calculation formula of the inverse distance weighted interpolation method is selected as follows: Weight , in, Z ( x 0 () represents the predicted value of pollutant concentration at unknown points. ，Z ( x j Let be the pollutant concentration at the j-th sampling point. d j For unknown points and the first j Euclidean distance of each sampling point ，p Distance attenuation coefficient ，The values ​​range from 1 to 3. For each of the directly and indirectly polluting environmental media within the study area, regional interpolation was performed for mercury, cadmium, arsenic, lead, and chromium to obtain the spatial distribution of pollution indices for different types of pollutants. Since the spatial interpolation is performed on the risk level of individual pollutants in different environmental media, rather than on the concentration of pollutants, the interpolation can reflect the pollution risk of individual pollutants in local areas.

[0042] The pollution local diffusion risk index is constructed by combining two parts: direct diffusion risk and indirect diffusion risk. Direct diffusion mainly includes the risk of heavy metals spreading directly into environmental media through natural processes such as soil erosion, surface runoff, groundwater migration, and sediment deposition. Indirect diffusion mainly refers to the risk of heavy metals spreading into agricultural products caused by human activities within and around the polluted area. The calculation method for the pollution local diffusion risk index is as follows: RDI 总 = RDI 直接 0.5+ RDI 间接 0.5, of which, RDI 直接 It serves as a diffusion risk index for soil and water media. RDI 间接 This is a diffusion risk index for agricultural products. RDI 总 The value classifies the risk of local spread into the following five levels: no spread RDI ≤1.0, slight diffusion 1.0< RDI ≤2.0, moderate diffusion 2.0< RDI ≤3.0, severe diffusion 3.0< RDI ≤4.0 and extremely severe diffusion RDI >4.0.

[0043] In this embodiment, the calculation formula for the sensitivity index model in step 4 is as follows: SI = A 1 ×5+ A 2 ×3+ A 3 ×1+ A 4 × (7+ N ),in SI Indicates the sensitivity index; A 1 、A 2 、A 3 、A 4These are the area percentages of main agricultural land, other agricultural land, residential land, and water area, respectively, and these percentages are calculated based on the total area of ​​the 2-kilometer buffer zone around the pollution source. N The number of water sources within the buffer zone is 7+N, which is the water area ratio coefficient. The sensitivity weight of water source protection is set with reference to HJ338-2018 "Technical Specification for Delineation of Drinking Water Source Protection Zones". N is the number of water sources within the buffer zone. For each additional water source, the sensitivity weight increases by 7 units, thus strengthening the priority of drinking water safety protection.

[0044] A sensitivity index dataset is formed by arranging the sensitivity indices of all pollution sources in ascending order. The 25%, 50%, 75%, and 90% quantiles of the sensitivity index dataset are then calculated using the linear interpolation formula described above. P 25 ´ , P 50 ´ , P 75 ´ , P 90 ´ .

[0045] according to P k ´ The quantile value divides the sensitivity index into five levels: Level I is extremely sensitive. SI ≥P 90 ´ Level II High Sensitivity P 75 ´ ≤ SI < P 90 ´ Level III, Moderately Sensitive P 50 ´ ≤ SI < P 75 ´ Level IV low sensitivity P 25 ´ ≤ SI < P 50 ´ Grade V Extremely Low Sensitivity SI < P 25 ´ .

[0046] In this embodiment, step 5 is determined based on regression analysis of historical monitoring data and the expert Delphi method. For this purpose, we invited 15 environmental experts and conducted three rounds of anonymous questionnaires, focusing on the impact of pollution source toxicity, volume, diffusion risk, and receptor sensitivity on actual risk events, obtaining importance scores for each indicator. Simultaneously, multiple linear regression was performed based on 100 sets of historical monitoring data, using the following weighting formula: To verify the correlation between each indicator and actual risk events, and in conjunction with expert opinions, the final weights were determined as follows: toxicity 40%, size 20%, local diffusion 25%, and remote receptor sensitivity 15%, while satisfying the following constraint: W1+W2+W3+W4=100%. The comprehensive risk value is calculated using the following formula: CR=W1·G1+W2·G2+W3·G3+W4·G4, where W1, W2, W3, and W4 represent the weights of the pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively; G1, G2, G3, and G4 represent the standardized scores of the pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively, with values ​​ranging from 0 to 10.

[0047] Based on the comprehensive risk value, priority control levels are divided into five levels: extremely high risk (CR≥8.0), high risk (6.0≤CR<8.0), medium risk (4.0≤CR<6.0), low risk (2.0≤CR<4.0), and extremely low risk (CR<2.0). The higher the priority control level, the greater the relative risk, and the higher the priority level should be included in the environmental management and risk control list.

[0048] In this embodiment, step 6, multi-source data fusion and model validation, specifically includes the following steps: Step 6.1 Data Acquisition and Preprocessing: Perform geometric correction, radiometric correction, noise removal, band fusion and color mosaic on the remote sensing images to ensure that the spatial resolution is not less than 1 meter; use the Z-score standardization method to eliminate dimensional differences in the sampled data; use linear interpolation to fill missing values ​​in the historical monitoring data and remove invalid data with a time span of more than 10 years. Step 6.2 Feature Extraction and Fusion: Extract normalized vegetation index, normalized water index and texture entropy features from remote sensing images; extract pollutant concentration gradient features and spatial clustering features from sampled data; extract annual average growth rate and seasonal fluctuation features of pollutant concentration from historical data, and construct a multidimensional feature matrix; Step 6.3 Spatiotemporal Joint Analysis Model Construction: Using spectral index, pollutant concentration gradient, and historical trend as input variables, the random forest algorithm is adopted, with the number of decision trees set to 100-500 and the maximum depth to 5-15 layers. Gini impurity is used as the node splitting criterion, and the parameters of splitting feature selection strategy, minimum leaf node sample number, and splitting threshold tolerance are optimized through cross-validation. Model outputs pollution diffusion probability P diffusion With receptor sensitivity score S sensitivty The overall risk weight is calculated using the following formula: , in, α These are the diffusion probability weighting coefficients. β The coefficients are the sensitivity score weights, and the coefficients satisfy the constraints. α + β =1, α The value range is 0.6-0.8. β The value range is 0.2-0.4; Step 6.4 Model Validation: Calculate the Pearson correlation coefficient, root mean square error, and [other parameters] using cross-validation. Ktuba The coefficients are used to verify the model's prediction accuracy and correct the risk level classification results. Specifically, the following indicators are used for model validation: The formula for calculating the Pearson correlation coefficient is: , in, x i For the first i The measured concentration values ​​of pollutants at each sampling point y i The model predicts the concentration values ​​for the corresponding points. , These are the average values ​​of the measured value and the predicted value, respectively. n The total number of samples is denoted as 1. The Pearson correlation coefficient is used to measure the degree of linear correlation between the model's predicted values ​​and the actual values. The closer the value is to 1, the higher the correlation between the model's predicted values ​​and the actual values.

[0049] The formula for calculating the root mean square error is: , in, x i For the first i The measured concentration values ​​of pollutants at each sampling point y i The model predicts the concentration value for the corresponding point. The root mean square error is used to quantify the degree of deviation between the model prediction value and the measured value. The smaller the value, the higher the accuracy of the model prediction.

[0050] Ktuba The formula for calculating the coefficient is: , in, N The total number of samples, m For the number of risk level categories, O kk For the first k The number of samples correctly classified into risk levels. R k and C k The first two results are the manually interpreted result and the model output result, respectively. k The total number of samples for each risk level Ktuba The closer the coefficient value is to 1, the higher the consistency between the model classification result and the human interpretation result.

[0051] Next, we will elaborate on the specific implementation process of the step-by-step environmental risk assessment method in complex multi-media and multi-scale scenarios to ensure that all technical features in the claims are fully supported by the specification.

[0052] This project focuses on a waste slag heap left over from a historical mining area in Hainan Province. This area poses a risk of heavy metal pollution, including cadmium, mercury, arsenic, lead, and chromium. Farmland, residential areas, and surface water bodies are distributed within a 2-kilometer radius. Environmental risk assessments need to be conducted on various environmental media, including soil, surface water, groundwater, and agricultural products, at different spatial scales, such as a local area of ​​500 meters and a remote area of ​​2 kilometers.

[0053] I. Basic Data Acquisition and Preprocessing Remote sensing image data: Satellite images with a resolution of 0.3-2 meters were collected from 2000 to 2022. After geometric correction, radiometric correction and band fusion processing, normalized vegetation index and normalized water index features were extracted for waste pile boundary identification and land use type classification.

[0054] On-site sampling data: For waste residue piles with an area >100m², sampling points were arranged using a 50m×50m grid method. A total of 50 samples of waste residue leachate, 80 soil samples, 60 agricultural product samples, 40 surface water samples, and 30 sediment samples were collected. Sampling points were evenly distributed on the surface of the waste residue pile, while soil sampling points were set at different distances around the waste residue pile. Samples of waste residue leachate, soil, agricultural products, surface water, and sediment were collected to detect heavy metal concentrations. The test results are as follows: the Cd concentration in the waste residue leachate was 150μg / L, and the Hg concentration was 50μg / L; the Cd concentration in the surrounding soil was 25mg / kg, and the Cd concentration in rice grains was 0.3mg / kg.

[0055] Historical monitoring data: Soil and surface water heavy metal concentration data from 2013 to 2022 were compiled, missing values ​​were filled in using linear interpolation, invalid data with a time span of more than 10 years were removed, and the average annual growth rate and seasonal fluctuation characteristics were extracted.

[0056] II. Toxicity Assessment of Pollution Sources Single-item pollution index method: Referring to the screening values ​​for construction land in GB36600-2018, Cd is taken as 50 μg / L and Hg as 0.1 μg / L. The calculation formula is as follows: PI was calculated Cd =3.0, indicating moderate Cd pollution, level 4; PI Hg =500; Hg is heavily polluted, level 5. Therefore, the single pollution risk level of this area is taken as the highest value of level 5.

[0057] Potential Ecological Risk Index Method: Based on the calculation formula... Based on GB36600-2018 and toxicological studies, the toxicity coefficients were selected as Cd=30 and Hg=40, i.e., RI=20090 in this case, far exceeding the "extremely high hazard" threshold, and the comprehensive ecological risk level was Level 1. Finally, based on the higher value of the individual and comprehensive toxicity levels of the pollution source, the toxicity of the pollution source in this area was determined to be Level 5, severe pollution.

[0058] III. Pollution Source Scale Assessment We statistically analyzed the volumes of 100 waste piles within the region, sorted them in ascending order, and calculated the 25th, 50th, 75th, and 90th percentile values, i.e., P. 25 =200m 3 P 50 =500m 3 P 75 =1000m 3 P 90 =2000m 3 The threshold was determined using a linear interpolation formula. The target waste pile volume in this area is 1500 m³. 3 Located at P 75 -P 90 The range is classified as "relatively large," indicating a high potential for pollutant release.

[0059] IV. Risk Assessment of Local Spread We used the inverse distance weighted interpolation method to analyze the spatial distribution of pollutants in soil, water bodies, and agricultural products, and calculated the diffusion risk index: Weight Take the distance attenuation coefficient p =2, generate a 50m resolution pollution distribution map, such as Picture 3As shown in Figure (a), which illustrates the direct diffusion PI value and pollution level of the area, and Figure (b), which illustrates the indirect diffusion PI value and pollution level of the area, it can be identified that the area with high direct diffusion risk accounts for 20%. Direct diffusion risk and indirect diffusion risk each account for 50% of the weight. This is based on the calculation method of the Local Diffusion Risk Index (RDI). 总 =RDI 直接 0.5+RDI 间接 0.5, direct diffusion RDI 直接 =3.6, Indirect Diffusion RDI 间接 =3.2, calculate RDI 总 =3.4, which shows that the area is moderately diffused, corresponding to level 3.

[0060] V. Remote Receptor Sensitivity Assessment Analyze land use types and water source distribution within a 2-kilometer radius, construct a buffer zone sensitivity index model, and calculate the sensitivity index using the following formula: SI = A 1 ×5+ A 2 ×3+ A 3 ×1+ A 4 × (7+ N (), where A1 is 40%, A2 is 20%, and N represents the number of water sources, which is 3. The calculation yields... SI =4. All pollution sources SI The values ​​are sorted, and the target is determined after calculating the quantile values. SI Located at P 50 ´-P 75 The interval was determined to be of moderate sensitivity, classified as level 3, indicating a moderate risk of remote receptor exposure.

[0061] VI. Comprehensive Risk Assessment Based on regression analysis of historical monitoring data and the Delphi method used by experts, the weights for toxicity (40%), size (20%), local diffusion (25%), and sensitivity (15%) were determined. Mapping each level yielded: toxicity level 5 → 10 points, size level 4 → 8 points, local diffusion level 3 → 6 points, and sensitivity level 3 → 6 points. According to the formula CR = 0.4 × 10 + 0.2 × 8 + 0.25 × 6 + 0.15 × 6 = 8.0, corresponding to "extremely high risk," and classified as Priority Control Level I. Therefore, it is currently recommended that risk control measures be implemented immediately for the waste slag heap left over from this historical mining area.

[0062] VII. Multi-source data fusion and model validation After geometric correction, the resolution of remote sensing images is ≤1 meter. Sampling data is standardized using Z-score, and missing values ​​are filled and invalid data are removed from historical data. Three-dimensional features such as NDVI, NWI, and texture entropy are extracted from remote sensing images. Two-dimensional features such as concentration gradient and spatial clustering are extracted from sampling data, and two-dimensional features such as annual growth rate and seasonal fluctuations are extracted from historical data, constructing a 7-dimensional feature matrix. The model is built using a random forest algorithm with 300 decision trees, a maximum depth of 10 layers, and a grid search optimization of node splitting criteria. The minimum number of leaf node samples is 5, the splitting threshold tolerance is 0.01, the output pollution diffusion probability is 0.85, and the receptor sensitivity score is 4.0. The comprehensive risk weight W = 0.7 × 0.85 + 0.3 × 4.0 = 1.795 is calculated. Cross-validation is used to calculate... R =0.91、 RMSE =0.9 and Ktuba =0.82, which confirms that the model has high prediction accuracy and the risk assessment results are reliable.

[0063] During the model validation process, the pollution diffusion probability of 0.85 output by the model indicates that the existing waste volume has a relatively high impact on the risk. There is a significant positive correlation between the pollution diffusion probability and the waste volume, given the volume of 1500 m³. 3 The waste pile has been assessed as extremely high-risk, therefore its management priority needs to be raised. Accordingly, we have increased the original 90th percentile from 2000m. 3 Adjusted to 1800m 3 Regarding the optimization of the distance attenuation coefficient, by comparing different... p The interpolation precision of the value, when p When the values ​​are 1, 2, and 3, the root mean square errors are 1.2, 0.9, and 1.1, respectively, and the final value is determined. p =2 is the optimal parameter. Therefore, the model with the optimized distance attenuation coefficient can more accurately predict the pollution diffusion range, thus guiding the development of more effective environmental remediation plans. For example... Picture 4 As shown, this environmental risk heat map has removed satellite map information, retaining only latitude and longitude markings. Latitude and longitude are used to assist in locating the geographical location of risk areas. Different colored areas represent different environmental risk levels; for example, red represents extremely high risk (CR≥8.0), and yellow represents high risk (6.0≤CR<8.0). This map visually demonstrates the differences in environmental risk distribution across different latitude and longitude regions, helping relevant departments and personnel quickly locate high-risk areas, rationally allocate environmental management resources, and formulate scientific and effective environmental risk control plans.

[0064] In summary, this method accurately identified the waste pile as a priority Class I pollution source, and the assessment results highly matched those obtained from rapid on-site detection and historical image interpretation. Model validation indicators demonstrate that this method can be applied to risk assessment and management of historical mining areas.

[0065] The foregoing has shown and described the main features and advantages of the present invention. It will be apparent to those skilled in the art that the specific embodiments of the present invention are not limited to the details of the exemplary embodiments described above, and various changes and improvements can be made without departing from the spirit or essential characteristics of the present invention. For example, new variables, such as wind speed and precipitation patterns, can be added to the feature matrix to further improve the model's adaptability to changes in climate factors; or the parameters of the random forest algorithm can be fine-tuned to adapt to differences in the environmental characteristics of specific regions. Furthermore, geographic information system technology can be combined to achieve the visualization of risk assessment results, assisting decision-makers in more efficient resource allocation and risk intervention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the present invention is defined by the appended claims rather than the foregoing description, and therefore all changes falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention.

[0066] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution, but may contain a combination of multiple technical solutions, which may be independent of each other or related to each other. For example, when dealing with multiple pollution sources, the risk of each pollution source can be assessed separately, and then the results can be integrated to achieve comprehensive management of pollution risk in the entire area. This approach not only enables refined control but also provides more comprehensive data support for the formulation of environmental policies. In this way, we can ensure the effectiveness and adaptability of environmental risk assessment and management, thereby protecting the ecological environment and promoting sustainable development. Furthermore, to improve the generalization ability of the model under different environmental conditions, cross-validation techniques from machine learning can be introduced. This narrative style of the specification is merely for clarity; those skilled in the art should consider the specification as a whole, and the technical solutions in the various embodiments can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for hierarchical assessment of environmental risk in a multi-medium multi-scale complex scene, characterized in that, Includes the following steps: Step 1: Toxicity assessment of pollution sources: Heavy metal concentration data in waste residue are obtained through leaching experiments. The individual pollution risk level is calculated using the single pollution index method and the comprehensive ecological risk level is assessed using the potential ecological risk index method. The toxicity level of the pollution source is determined based on the higher value of the two risk levels. Step 2: Pollution source size assessment: Based on the quantile values ​​of pollution source size, the size levels are classified to identify pollution sources that require priority control. Step 3: Local diffusion risk assessment: Spatial interpolation of pollutant concentrations in soil, water bodies, and agricultural products is performed using the inverse distance weighted interpolation method. The local diffusion risk level is then determined by combining the weighted calculation results of the direct diffusion risk index and the indirect diffusion risk index. Step 4: Remote diffusion receptor sensitivity assessment: Analyze the land use types and water source distribution within a 2-kilometer buffer zone around the pollution source to construct a sensitivity index model. Calculate the sensitivity index of all pollution sources using quantile values ​​and classify the remote receptor sensitivity levels. Step 5: Comprehensive Risk Assessment: The toxicity level, volume level, local diffusion risk level, and remote receptor sensitivity level obtained from Steps 1 to 4 are weighted and summed according to preset weights to generate the pollution source priority control level. Step 6: Multi-source data fusion and model validation: Integrate multi-temporal high-resolution remote sensing images, field sampling data, and historical environmental monitoring data to construct a spatiotemporal joint analysis model and validate the reliability of the risk assessment results; Step 6, multi-source data fusion and model validation, specifically includes the following steps: Step 6.1 Data Acquisition and Preprocessing: Perform geometric correction, radiometric correction, noise removal, band fusion and color mosaic on the remote sensing images to ensure that the spatial resolution is not less than 1 meter; use the Z-score standardization method to eliminate dimensional differences in the sampled data; use linear interpolation to fill missing values ​​in the historical monitoring data and remove invalid data with a time span of more than 10 years. Step 6.2 Feature Extraction and Fusion: Extract normalized vegetation index, normalized water index and texture entropy features from the preprocessed remote sensing image; extract pollutant concentration gradient features and spatial clustering features from the preprocessed sampling data; extract the annual average growth rate of pollutant concentration and seasonal fluctuation features from the preprocessed historical data, and construct a multidimensional feature matrix; Step 6.3 Spatiotemporal Joint Analysis Model Construction: Using the multidimensional feature matrix as the input variable, the random forest algorithm is adopted, the number of decision trees is set to 100-500, the maximum depth is 5-15 layers, the Gini impurity is used as the node splitting criterion, and the parameters of splitting feature selection strategy, minimum number of leaf node samples and splitting threshold tolerance are optimized through cross-validation. Model outputs a pollution dispersion probability P diffusion With the receptor sensitivity score S sensitivity and the integrated risk weight is calculated by the following formula: ， wherein, α is a diffusion probability weight coefficient, β is a sensitivity score weight coefficient, and the coefficients satisfy the constraint condition α + β = 1, α has a value range of 0.6-0.8, β has a value range of 0.2-0.4; Step 6.4 Model validation: Calculate the Pearson correlation coefficient, root mean square error and Kappa coefficient, validate the prediction accuracy of the model and revise the risk grade classification result; The output of the spatiotemporal joint analysis model in step 6 is used to dynamically update the threshold for classifying pollution source volume levels, optimize the distance attenuation coefficient in the inverse distance weighted interpolation method, generate a risk heat map, and identify priority control levels.

2. The method according to claim 1, characterized in that, In step 1, according to PI i The risk level of individual pollution items is divided and mapped to five levels. The calculation formula for the individual pollution index method is as follows: ,in, PI i For a single pollution index, C i For the first i The measured concentrations of several heavy metals, S i For the first i Environmental quality standard limits for each type of heavy metal; In step 1, according to RI The comprehensive ecological risk level is divided into five levels, and the calculation formula for the potential ecological risk index method is as follows: ,in In the formula: For the first i Individual potential risk indices for each heavy metal For the first i The toxicity coefficient of each heavy metal; The higher value between the individual pollution risk level and the comprehensive ecological risk level is used as the toxicity level of the pollution source. ,in, L PI , L RI These are numerical levels of 1-5, representing both individual pollution risk and comprehensive ecological risk, with higher values ​​indicating higher risk.

3. The method according to claim 2, characterized in that, The specific method for classifying pollution source volume levels based on quantile values ​​in step 2 is as follows: Arrange the waste residue volumes of all blocks in ascending order to form a pollution source volume dataset. Calculate the 25%, 50%, 75%, and 90% quantile values ​​of the pollution source volume dataset. Based on the quantile values, classify the pollution source volume into five levels. The quantile values ​​are calculated using the linear interpolation formula: ,in, n The total number of sampling points. k For quantiles, X This is the sorted dataset.

4. The method according to claim 1, characterized in that, The calculation formula for the inverse distance weighted interpolation method in step 3 is as follows: Weight , in, Z ( x 0 () represents the predicted value of pollutant concentration at unknown points. ，Z ( x j Let be the pollutant concentration at the j-th sampling point. d j For unknown points and the first j Euclidean distance of each sampling point ，p Distance attenuation coefficient ， The value ranges from 1 to 3; n This represents the total number of sampling points.

5. The method according to claim 4, characterized in that, In step 3, according to RDI 总 The risk of local spread is divided into five levels, and the formula for calculating the comprehensive spread risk index is as follows: RDI 总 = RDI 直接 0.5+ RDI 间接 0.5, of which, RDI 直接 It serves as a diffusion risk index for soil and water media. RDI 间接 This is a diffusion risk index for agricultural products.

6. The method according to claim 3, characterized in that, The formula for calculating the sensitivity index model in step 4 is as follows: SI = A 1 ×5+ A 2 ×3+ A 3 ×1+ A 4 ×（7+ N ）， in SI Indicates the sensitivity index; A 1 、A 2 、A 3 、A 4 These are the area percentages of main agricultural land, other agricultural land, residential land, and water area, respectively, and these percentages are calculated based on the total area of ​​the 2-kilometer buffer zone around the pollution source. N 7+N represents the number of water sources within the buffer zone, and 7+N is the water area ratio coefficient. Arrange the sensitivity indices of all pollution sources in ascending order to form a sensitivity index dataset. Calculate the 25%, 50%, 75%, and 90% quantiles of the sensitivity index dataset using the linear interpolation formula in step 2. Divide the sensitivity index into five levels based on the quantiles.

7. The method according to claim 6, characterized in that, In step 5, the weighting rule is as follows: 40% for pollution source toxicity level, 20% for volume level, 25% for local diffusion risk level, and 15% for remote receptor sensitivity level. A comprehensive risk value is obtained by weighted summation. Based on the comprehensive risk value, the priority control levels are divided into five levels. The formula for calculating the comprehensive risk value is as follows: CR=W1·G1+W2·G2+W3·G3+W4·G4, Wherein, W1, W2, W3, and W4 represent the weights of the pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively; G1, G2, G3, and G4 represent the standardized score values ​​of the pollution source toxicity level, size level, local diffusion risk level, and remote receptor sensitivity level, respectively, with values ​​ranging from 0 to 10.

Images