A method for dynamic assessment and early warning of heavy metal pollution in river and lake sediment

Abstract

The present application relates to the technical field of environmental monitoring and pollution assessment, and particularly relates to a risk dynamic assessment and early warning method for heavy metal pollution of river and lake sediment, which comprises collecting sediment samples and water ecological data, analyzing heavy metal form distribution and classification, simulating environmental dynamic fluctuation to assess form stability, identifying high-risk forms in combination with particle size and organic matter content, constructing a Bayesian network model to calculate release risk value, and optimizing the model through measured data, and finally realizing dynamic assessment and automatic early warning. The present application can improve the accuracy and response timeliness of heavy metal pollution risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring and pollution assessment technology, and specifically relates to a method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments. Background Technology

[0002] In the fields of environmental science and pollution control, studying the distribution and behavior of heavy metals in aquatic sediments is of paramount importance, directly impacting ecosystem health and human safety. Sediment serves as a primary deposition site for pollutants, and the heavy metals within it have a profound impact on water quality and the food chain. Especially when environmental conditions change, they may trigger the re-release of pollutants, threatening environmental safety. Therefore, in-depth investigation into the state of heavy metals in sediments and their potential risks has become a crucial issue that cannot be ignored in ensuring ecological balance and public health.

[0003] Previous studies have largely focused on total heavy metal measurement, with less consideration given to the variability in their occurrence formats, leading to biased assessments of environmental behavior. Current technologies lack coupled modeling of dynamic changes in aquatic ecological factors such as pH, dissolved oxygen, and temperature in heavy metal speciation and environmental response analysis, limiting the accurate characterization of speciation processes. Furthermore, when assessing the impact of sediment particle size and organic matter content on heavy metal adsorption-desorption kinetics, quantitative models are typically not established, making it difficult to support high-precision risk prediction. In practical applications, the lack of a model optimization mechanism based on measured data feedback often results in discrepancies between release predictions and on-site monitoring results, affecting the timeliness and reliability of risk warnings. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments, comprising the following steps: Step S1: Collect sediment sample data from the target water body to obtain information on the distribution of heavy metal speciation. Combine the sediment particle size, organic matter content, and water body ecological conditions data to construct an initial dataset for subsequent analysis. The water body ecological conditions data include pH, temperature, and oxygen content. Step S2: Preprocess the distribution information of heavy metal speciation in the initial dataset, use a classification method to distinguish the chemical state change characteristics of different speciations, and obtain the classified speciation dataset. The classified speciation dataset is used for stability assessment. Step S3: Based on the classified morphological dataset, obtain environmental change rate data, and analyze the stability of each morphology under different conditions by simulating the dynamic fluctuations of aquatic ecological conditions, and determine the trend of stability change. Step S4: Extract the impact data of sediment particle size and organic matter content on the heavy metal adsorption and release rate from the stability change trend, determine whether the adsorption and release rate exceeds the preset threshold, and if it does, mark it as a high-risk form and generate a risk form identifier set. Step S5: For the risk form identifier set, combine the environmental change rate and chemical state change data to construct a release risk value assessment model, calculate the potential release probability of each form under simulated environmental dynamic fluctuation conditions, and determine the release risk value distribution. Step S6: Based on the distribution of release risk values, obtain prediction deviation data. By comparing the difference between the release amount under simulated conditions and the actual monitoring data, adjust the parameters of the release risk value assessment model to obtain the optimized risk prediction results. Step S7: Using the optimized risk prediction results, generate a dynamic monitoring plan for high-risk patterns, continuously update the pattern stability and risk value data, and trigger an early warning mechanism through automatic notification and emergency response measures if the risk value exceeds the safe range.

[0006] In some embodiments, the heavy metal speciation information includes the proportions of exchangeable, carbonate-bound, iron-manganese oxide-bound, organically bound, and residual forms; the classified speciation dataset is categorized by chemical extraction method and its reactivity level is labeled; the stability change trend includes the transformation path map of each speciation under pH gradient, dissolved oxygen fluctuation, and temperature change; the risk speciation identifier set includes speciation type, corresponding sediment physicochemical parameters, and critical release rate threshold; the release risk assessment model uses a Bayesian network structure to integrate multivariate inputs; the prediction deviation data quantifies the deviation between the model output and the measured release amount using the root mean square error index; the dynamic monitoring scheme includes adaptive adjustment rules for sampling frequency, sensor deployment location optimization strategy, and dynamic correction mechanism for risk threshold; the early warning mechanism pushes real-time alarms to the environmental management platform and triggers response procedures for water quality control equipment through message queue telemetry transmission protocol.

[0007] In some implementations, step S1, which involves collecting sediment sample data from the target water body, obtaining information on the distribution of heavy metal speciation, and constructing an initial dataset by combining sediment particle size, organic matter content, and water body ecological conditions data, specifically includes: Step S101: Set up gridded sampling points in the target river and lake, use a columnar sampler to obtain bottom sediment samples at a depth of 0–20 cm, and simultaneously deploy multi-parameter water quality probes to record the pH, dissolved oxygen concentration and water temperature at the sampling time to form a spatiotemporally matched original observation record. Step S102: After freeze-drying the sediment sample, pass it through a 100-mesh sieve. Use BCR continuous extraction method to separate five heavy metal occurrences. Use inductively coupled plasma mass spectrometry to determine the concentrations of cadmium, lead, copper, zinc and chromium in each occurrence and generate a speciation matrix. Step S103: Take sediment samples from the same batch for laser particle size analysis to obtain particle size distribution curves, and use potassium dichromate oxidation-external heating method to determine organic carbon content, which is then converted into organic matter content index. Step S104: Structure the morphological distribution matrix, particle size data, organic matter content and water ecological parameters according to the sampling point number to construct an initial dataset containing spatial coordinates, timestamps and multidimensional attribute fields.

[0008] In some implementations, the step S2, which involves preprocessing the heavy metal speciation distribution information in the initial dataset and using a classification method to distinguish the chemical state change characteristics of different speciations to obtain a classified speciation dataset, specifically includes: Step S201: Normalize the concentration of each morphology in the initial dataset to eliminate dimensional differences, and use principal component analysis to extract the main variation direction of morphological combinations; Step S202: Based on the principal component scores, the K-means clustering algorithm is applied to divide the heavy metal forms into a highly active group (exchangeable state + carbonate-bound state), a moderately active group (iron-manganese oxide-bound state + organic-bound state), and an inert group (residue state), and each group is assigned a stability label; Step S203: Introduce the fuzzy C-means algorithm to assign membership degrees to boundary samples, avoid information loss caused by hard partitioning, and generate morphological classification results with fuzzy membership degrees; Step S204: Bind the classification results with the physicochemical parameters of the sediment to form a classified morphological dataset containing morphological categories, membership coefficients, and corresponding environmental factors.

[0009] In some implementations, in step S3, the step of obtaining environmental change rate data based on the classified morphological dataset, analyzing the stability of each morphology under different conditions by simulating the dynamic fluctuations of aquatic ecological conditions, and determining the trend of stability change specifically involves: S301: Construct an environmental disturbance simulation chamber, set a three-factor orthogonal experimental matrix with pH 5.0–9.0, dissolved oxygen 0–12 mg / L, and temperature 5–35℃, and place the classified sediment samples in the chamber for dynamic exposure; Step S302: Collect overlying water samples every N1=24 hours, determine the concentration of dissolved heavy metals, calculate the release flux per unit time, and generate a time-series release curve; S303: Perform slope analysis on the release curve, identify the inflection point of the release rate of each form under different environmental combinations, and establish a mapping relationship table between the environmental change rate and the release rate. Step S304: Use grey relational analysis to quantify the influence weight of each environmental factor on the stability of different forms, and draw a heat map of the stability change trend.

[0010] In some implementations, in step S4, the step of extracting the influence data of sediment particle size and organic matter content on the heavy metal adsorption and release rate from the stability change trend, determining whether the adsorption and release rate exceeds a preset threshold, and marking it as a high-risk form if it does, and generating a risk form identifier set, specifically includes: Step S401: Based on the heat map of stability change trend, select the experimental group with release rate greater than the first release rate threshold of 0.05 μg / (g·d) and extract the corresponding median particle size D50 and organic matter content of the sediment. Step S402: Establish a multivariate nonlinear regression model with D50 and organic matter content as independent variables and release rate as dependent variable, and fit the adsorption-desorption kinetic response surface. Step S403: Set the second release rate threshold to 0.08 μg / (g·d). When the model prediction value exceeds this threshold, the corresponding morphology is marked as high risk. Step S404: Encapsulate the type of high-risk form, its active group, critical D50 range, organic matter content range, and threshold parameters into structured entries to generate a risk form identifier set.

[0011] In some implementations, in step S5, the step of constructing a release risk value assessment model for the risk form identifier set, combining environmental change rate and chemical state change data, calculating the potential release probability of each form under simulated environmental dynamic fluctuation conditions, and determining the release risk value distribution specifically includes: Step S501: Using a Bayesian network as a framework, define nodes including environmental change rate, sediment physicochemical parameters, and morphological categories, and edges to represent causal dependencies; Step S502: Use historical experimental data to train a conditional probability table and determine the release probability distribution of child nodes under each combination of parent nodes; Step S503: Input real-time or predicted aquatic ecological parameters and infer the release probability of each risk form through belief propagation algorithm; Step S504: Multiply the release probability by the morphological toxicity weight, sum the weighted values ​​to obtain the comprehensive release risk value, and generate a spatialized risk value distribution map.

[0012] In some implementations, in step S6, the step of obtaining prediction deviation data based on the release risk value distribution, adjusting the parameters of the release risk value assessment model by comparing the release amount under simulated conditions with the actual monitoring data, and obtaining the optimized risk prediction result specifically involves: Step S601: Deploy an in-situ release monitoring device in the target water area, record the changes in heavy metal concentration in the overlying water for 30 consecutive days, and calculate the measured cumulative release amount; Step S602: Run the release risk value assessment model, input the environmental parameters for the same period, and output the simulated release amount; Step S603: Calculate the root mean square error (RMSE) between the simulated value and the measured value. If the RMSE > the first error threshold of 0.15, then trigger the parameter correction process. Step S604: Use the Levenberg-Marquardt algorithm to invert and adjust the conditional probability parameters in the Bayesian network until RMSE ≤ the second error threshold of 0.10, then output the optimized risk prediction result.

[0013] In some implementations, in step S7, the step of generating a dynamic monitoring scheme for high-risk patterns using the optimized risk prediction results, continuously updating pattern stability and risk value data, and triggering an early warning mechanism through automatic notification and emergency response measures if the risk value exceeds the safe range, specifically includes: Step S701: Set the risk value safety threshold. When the output value of the optimized model exceeds the threshold, automatically increase the sampling frequency of the corresponding area to once a week. Step S702: Deploy an IoT water quality sensor array in the high-risk area to transmit pH, DO, temperature and heavy metal ion concentration data in real time; Step S703: Establish a risk value sliding window update mechanism, and refresh the prediction results based on new data every N2=24 hours; Step S704: When two consecutive update values ​​exceed the threshold, a structured alarm is sent to the environmental monitoring platform, and an emergency response procedure for aeration or addition of passivating agents is initiated.

[0014] Compared with the prior art, the present invention has the following beneficial effects: To address the shortcomings of existing methods that focus solely on total heavy metal content while neglecting speciation differences, lack coupling analysis with dynamic environmental changes, and fail to establish a quantitative relationship between sediment physicochemical properties and release behavior, this invention systematically collects and analyzes the distribution of heavy metal speciation. By combining sediment particle size, organic matter content, and aquatic ecological parameters, it achieves precise identification of the occurrence state of heavy metals in river and lake sediments. Through simulating dynamic environmental fluctuations, a Bayesian network-based release risk assessment model is constructed to quantify the release probability and risk distribution of different speciations under changing conditions. Furthermore, by continuously optimizing model parameters using measured data, a closed-loop mechanism of monitoring, assessment, early warning, and response is formed, significantly improving the accuracy and timeliness of risk prediction and providing dynamic and reliable technical support for the prevention and control of heavy metal pollution in aquatic environments. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but this is not to limit the scope of the invention to this.

[0017] like Figure 1 As shown, the present invention provides a method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments, which may specifically include the following steps: Step S1: Collect sediment sample data from the target water body to obtain information on the distribution of heavy metal speciation. Combine the sediment particle size, organic matter content, and water body ecological conditions data to construct an initial dataset. The water body ecological conditions data include pH, temperature, and oxygen content.

[0018] Specifically, a grid of sampling points was first established in the target river and lake area. The grid spacing was set at 500 m × 500 m based on the water area and historical pollution distribution. At each sampling point, a columnar sampler was vertically inserted into the sediment to a depth of 20 cm to obtain undisturbed columnar sediment samples. Simultaneously, a YSI ProDSS multi-parameter water quality probe was deployed at 0.5 m below the water surface at the sampling points to record the pH, dissolved oxygen (DO), and water temperature (T) at the time of sampling, forming a raw observation record with a unified timestamp and spatial coordinates. The collected sediment samples were immediately frozen at -20℃ and transported back to the laboratory for freeze-drying to remove moisture. After sieving, the samples were passed through a 100-mesh standard sieve to eliminate test bias caused by particle unevenness.

[0019] Subsequently, a five-step chemical separation process using the BCR sequential extraction method was employed to obtain five heavy metal occurrence forms: exchangeable, carbonate-bound, iron-manganese oxide-bound, organically bound, and residual. The concentrations of cadmium (Cd), lead (Pb), copper (Cu), zinc (Zn), and chromium (Cr) in each form were determined using inductively coupled plasma mass spectrometry (ICP-MS), generating a speciation matrix with sampling point numbers as rows, speciation categories as columns, and element concentrations as units. Simultaneously, sediment samples from the same batch were sent to a Malvern laser particle size analyzer (Mastersizer 3000) to obtain particle size distribution curves and calculate the median particle size (D50). A portion of the samples were then subjected to potassium dichromate oxidation-external heating to determine the organic carbon content, which was then multiplied by a conversion factor of 1.724 to obtain the organic matter content index.

[0020] All the above data were structured and integrated according to the sampling point number to form an initial dataset containing 28 fields, including spatial coordinates (longitude and latitude), timestamp, pH, DO, T, D50, organic matter content, and concentration of five heavy metals in five forms. This dataset serves as the input for subsequent form classification.

[0021] Step S2: Preprocess the distribution information of heavy metal speciation in the initial dataset, use a classification method to distinguish the chemical state change characteristics of different speciations, and obtain the classified speciation dataset.

[0022] Specifically, firstly, the concentrations of each morphology in the initial dataset were subjected to Min-Max normalization, compressing the values ​​to the [0,1] interval to eliminate dimensional differences. Then, Principal Component Analysis (PCA) was performed on the normalized data to extract principal components with a cumulative variance contribution rate exceeding 85%, thus obtaining the main variation directions of the morphological combinations. Finally, based on the principal component scores, the K-means clustering algorithm was used to divide all samples into three groups: The first group contains exchangeable and carbonate-bound states and is defined as the highly active group. The second group includes both iron-manganese oxide bound state and organic bound state, and is defined as the medium-active group. The third group is the residue state, defined as the inert group.

[0023] To avoid the loss of boundary sample information due to hard clustering, a fuzzy C-means (FCM) algorithm is further introduced. This algorithm calculates the membership coefficient of samples at the cluster boundaries to each group. For example, a sample might have a membership coefficient of 0.62 to the highly active group, 0.35 to the moderately active group, and 0.03 to the inert group. Finally, the morphological category and membership coefficient of each sample are bound to its corresponding sediment physicochemical parameters (D50, organic matter content) and aquatic ecological parameters (pH, DO, T) to form a classified morphological dataset for subsequent stability assessment.

[0024] Step S3: Based on the classified morphological dataset, obtain environmental change rate data, and analyze the stability of each morphology under different conditions by simulating the dynamic fluctuations of aquatic ecological conditions, and determine the trend of stability change.

[0025] Specifically, a closed environmental disturbance simulation chamber was first constructed. The chamber was made of corrosion-resistant polytetrafluoroethylene (PTFE) material, with an internal volume of 50 L. It was equipped with an automatic pH adjustment system, a dissolved oxygen control system, and a constant-temperature circulating water bath. Classified sediment samples were placed into dialysis bags (molecular weight cutoff 10 kDa) according to their activity group and placed at the bottom of the simulation chamber, covered with 30 L of deionized water. pH gradients of 5.0, 6.5, 8.0, and 9.0, dissolved oxygen concentrations of 0, 4, 8, and 12 mg / L, and temperatures of 5℃, 15℃, 25℃, and 35℃ were set, forming 4×4×4=64 sets of orthogonal experimental conditions. Each experiment lasted for 7 days. Every 24 hours, 10 mL of the covered water sample was extracted using a 0.45 μm filter membrane. After acidification with nitric acid, the concentrations of dissolved Cd, Pb, Cu, Zn, and Cr were determined using ICP-MS. The heavy metal release flux per unit mass of sediment per unit time (μg / (g·d)) was calculated, generating a time-series release curve.

[0026] Then, the first derivative of each curve is calculated to identify the inflection point time when the release rate changes significantly, and a mapping table between the combination of environmental factors (pH, DO, T) and the release rate is established.

[0027] Next, grey relational analysis was used, with release rate as the reference sequence and each environmental factor as the comparison sequence, to calculate the correlation coefficient and average it, thus obtaining the weight of each factor on the stability of different morphologies. For example, pH has a weight of 0.82 for the exchangeable state, and DO has a weight of 0.76 for the bound state of iron and manganese oxides. The weighting results are then visualized as a heatmap, with the horizontal axis representing morphological categories and the vertical axis representing environmental factors, and the color intensity indicating the strength of the influence.

[0028] Step S4: Extract the impact data of sediment particle size and organic matter content on the adsorption and release rate of heavy metals from the stability change trend, determine whether the adsorption and release rate exceeds the preset threshold, and if it does, mark it as a high-risk form and generate a risk form identifier set.

[0029] Specifically, based on the thermogram output in step S3, all experimental groups with release rates greater than 0.05 μg / (g·d) were selected, and their corresponding sediment D50 values ​​and organic matter content were extracted. Subsequently, a multivariate nonlinear regression model was established, with D50 (unit: μm) and organic matter content (unit: %) as independent variables and release rate (unit: μg / (g·d)) as the dependent variable. Gaussian process regression was used to fit the adsorption-desorption kinetic response surface.

[0030] An engineering safety threshold of 0.08 μg / (g·d) is set. When the release rate predicted by the model exceeds this value, the corresponding form is identified as a high-risk form. For example, when D50 < 10 μm and organic matter content < 2%, the predicted release rate of exchangeable Cd reaches 0.092 μg / (g·d), exceeding the threshold, and is therefore marked as high-risk. All high-risk forms are encapsulated into JSON-formatted structured entries, including their type (e.g., "exchangeable-Cd"), their corresponding activity group (high activity group), critical D50 range (e.g., 5–10 μm), organic matter content range (e.g., 1.5–2.0%), and threshold parameter (0.08 μg / (g·d)), forming a risk form identifier set.

[0031] Step S5: For the risk form identifier set, combine environmental change rate and chemical state change data to construct a release risk value assessment model, calculate the potential release probability of each form under simulated environmental dynamic fluctuation conditions, and determine the release risk value distribution.

[0032] Specifically, a Bayesian network is used as the modeling framework, defining nodes as follows: parent nodes "environmental change rate" (composed of pH change rate, DO change rate, and T change rate), "sediment physicochemical parameters" (D50, organic matter content), and "morphological categories" (five morphologies); child nodes are "release probabilities." Nodes are connected by directed edges, representing causal dependencies; for example, "pH change rate" points to "commutative release probability." A conditional probability table (CPT) is trained using historical experimental data accumulated in steps 3 and 4, and the maximum likelihood estimation method is used to determine the probability distribution of child nodes under each combination of parent nodes.

[0033] In actual operation, the current or short-term forecast of aquatic ecological parameters are input (such as the pH expected to drop from 7.2 to 6.0 and DO from 8.5 mg / L to 5.0 mg / L in the next 24 hours), and the probability of release of each risk form is output by performing probabilistic inference in the Bayesian network through the Belief Propagation algorithm.

[0034] Subsequently, toxicity weighting factors (e.g., Cd=1.5, Pb=1.2, Cu=1.0, Zn=0.8, Cr=1.1) are introduced. The release probability is multiplied by the toxicity weight and then summed in a weighted manner to obtain a comprehensive release risk value (range 0–1). Finally, the risk values ​​of each sampling point are used to generate a spatially continuous risk value distribution map through inverse distance weighted interpolation.

[0035] Step S6: Based on the distribution of release risk values, obtain prediction deviation data. By comparing the difference between the release amount under simulated conditions and the actual monitoring data, adjust the parameters of the release risk value assessment model to obtain the optimized risk prediction results.

[0036] Specifically, an in-situ release monitoring device is first deployed in the high-risk area of ​​the target water body. This device consists of a sediment trap, an overlying water sampling bottle, a micro-pump, and a data logger. It can automatically collect overlying water samples and measure heavy metal concentrations for 30 consecutive days, calculating the measured cumulative release. Simultaneously, the measured environmental parameters are input into the Bayesian network model in step S5 to output the simulated release.

[0037] Then, the root mean square error (RMSE) between the simulated and measured values ​​is calculated using the formula RMSE = √(Σ(y_i - ŷ_i)² / n), where y_i is the measured value, ŷ_i is the simulated value, and n is the number of sampling days. If RMSE > 0.15, the parameter calibration process is initiated.

[0038] Specifically, the Levenberg-Marquardt nonlinear least squares algorithm is used, with RMSE as the objective function, to inversely adjust the parameters of the conditional probability table in the Bayesian network, iteratively updating until RMSE ≤ 0.10. The optimized conditional probability table replaces the original model parameters, outputting the optimized risk prediction results.

[0039] Step S7: Using the optimized risk prediction results, generate a dynamic monitoring plan for high-risk patterns, continuously update the pattern stability and risk value data, and trigger an early warning mechanism through automatic notification and emergency response measures if the risk value exceeds the safe range.

[0040] Specifically, a safety threshold for the overall risk value is first set to 0.65. When the risk value output in step S6 exceeds this threshold, the system automatically increases the sampling frequency of the corresponding area from once a month to once a week, and deploys an IoT water quality sensor array in the area, including a pH sensor (model: PH-801), a dissolved oxygen sensor (model: DO-2000), a temperature sensor (model: PT100), and a heavy metal ion selective electrode (for Cd²⁺ and Pb²⁺). All sensors upload data to the cloud server in real time via a LoRa wireless module.

[0041] Simultaneously, a 24-hour sliding window mechanism is established. Every 24 hours, after receiving new data, the optimized Bayesian network model is re-run to refresh the risk prediction results. If the risk value exceeds 0.65 in two consecutive updates, an early warning mechanism is triggered. Structured alarm messages are pushed to the environmental monitoring platform via the MQTT protocol. The message content includes fields such as location coordinates, the form of exceeding the standard, risk value, and recommended measures. At the same time, it can also link with the water quality control equipment deployed in the water area to automatically start the microporous aeration device to increase dissolved oxygen, or spray phosphate passivating agents (such as hydroxyapatite) onto the surface of the bottom sediment through a dosing pump to inhibit the release of heavy metals.

[0042] This embodiment forms a closed-loop control system from data acquisition, analysis, evaluation to response, with each step closely connected to ensure the dynamism of the evaluation results and the timeliness of the early warning.

[0043] To enable those skilled in the art to fully understand and implement this invention, the following further supplements the specific implementation principle of this invention in the application scenario of sediment pollution assessment in an urban lake.

[0044] Step 1: Based on the lake's area of ​​approximately 6.5 km² and historical heavy metal monitoring hotspots, 169 grid-like sampling points (13×13 grids) were established on the lake surface, spaced 500 m × 500 m apart. At each sampling point, a stainless steel cylindrical sampler was used to vertically penetrate the sediment to a depth of 20 cm to obtain undisturbed core samples. Simultaneously, a YSI ProDSS multi-parameter probe was deployed 0.5 m below the water surface to record ecological parameters such as pH=7.8, DO=6.2 mg / L, and T=22.3℃ in real time, assigning them uniform GPS coordinates and timestamps. The obtained sediment samples were immediately placed in pre-cooled aluminum foil bags and transported to the laboratory via a -20℃ cold chain. After freeze-drying, the sample was passed through a 100-mesh sieve to eliminate the influence of particle size heterogeneity on extraction efficiency. Subsequently, a five-step continuous extraction method using BCR was employed to separate exchangeable, carbonate-bound, iron-manganese oxide-bound, organic-bound, and residual forms of Cd. ICP-MS analysis revealed that the concentration of Cd in the exchangeable form was 1.82 mg / kg, significantly higher than other forms. Simultaneously, D50 = 8.3 μm and organic matter content = 1.7% were determined. Finally, the 28 parameters were structured and stored according to sampling point IDs to form an initial dataset, providing a high-fidelity input source for subsequent morphology classification.

[0045] Step 2: Min-Max normalization was performed on the speciation concentration matrix of the 169 samples from Step 1, mapping all element concentrations to the [0,1] interval to eliminate the dimensional differences between Cd (μg level) and Cr (mg level). Principal component analysis was then performed. The cumulative variance contribution rate of the first two principal components reached 89.3%, with PC1 mainly reflecting the covariance between exchangeable and carbonate-bound states, and PC2 characterizing the coupling characteristics of iron and manganese oxides with organic matter. Based on the principal component scores, K-means clustering divided the samples into three groups: Group 1 (highly active) contains 42 sites with exchangeable Cd content >60%; Group 2 (medium active) is dominated by Cu in iron-manganese oxide bound state; Group 3 (inert) is dominated by residual Cr.

[0046] For the 15 samples located at the boundary between Group 1 and Group 2 (e.g., a membership degree of 0.58 for exchangeable Cd and 0.41 for Cu bound to iron-manganese oxides at a certain location), the fuzzy C-means algorithm was used to calculate their membership vectors to each group, avoiding hard partitioning that would distort boundary information. The final output classification dataset not only includes morphological category labels but also embeds membership coefficients, D50, organic matter content, and original pH / DO / T values, ensuring that subsequent stability simulations can distinguish the response characteristics of different activity levels.

[0047] Step 3: Select typical samples (D50=8.3 μm, organic matter content=1.7%) from the high-activity group in Step 2, place them in dialysis bags with a molecular weight cutoff of 10 kDa, and put them at the bottom of a 50 L polytetrafluoroethylene simulation chamber, covered with 30 L of deionized water. The experimental conditions were set as follows: pH=6.5 (simulating an acid rain event), DO=4 mg / L (simulating summer hypoxia), and T=35℃ (simulating a heat wave). On the 3rd day of the experiment, the Cd concentration in the covered water suddenly increased to 12.4 μg / L, and the calculated release flux reached 0.091 μg / (g·d), far exceeding other conditions. Calculating the first derivative of the time-series release curve revealed an inflection point on day 2.8, indicating that acidic conditions accelerated the desorption kinetics of exchangeable Cd. Grey relational analysis showed that the correlation coefficient between pH and the release rate of exchangeable Cd was 0.82, significantly higher than that between DO (0.31) and T (0.45), indicating that acidity was the dominant factor. The weighting results are encoded into a heatmap matrix, with the horizontal axis representing five types of morphology and the vertical axis representing three types of environmental factors, with color intensity quantifying the degree of influence.

[0048] Step 4: Based on the heatmap matrix obtained in Step 3, extract all experimental groups with release rates > 0.05 μg / (g·d) from the heatmap, obtaining a total of 28 valid data sets. Taking the data point with D50 = 8.3 μm, organic matter content = 1.7%, and release rate = 0.091 μg / (g·d) as an example, it is included in the training set of the Gaussian process regression model. The model fitting results show that when D50 < 10 μm and organic matter content < 2%, the release rate surface of exchangeable Cd rises sharply, with a predicted value of 0.092 μg / (g·d), exceeding the engineering threshold of 0.08 μg / (g·d). Therefore, "exchangeable-Cd" is automatically labeled as a high-risk form, and its critical parameter range is recorded (D50: 5–10 μm; organic matter content: 1.5–2.0%). The structured entries are encapsulated in JSON format and include morphology type, activity group, threshold and physicochemical boundary, forming a risk morphology identifier set to provide accurate risk unit definition for subsequent applications.

[0049] Step 5: In the risk assessment model, construct a Bayesian network topology, where the "pH change rate" node is connected to the "commutative Cd release probability" child node via directed edges. Use the 64 sets of orthogonal experimental data accumulated in Steps 3 and 4 to train the conditional probability table. For example, when the pH change rate is < -0.5 units / 24h, D50 = 8.3 μm, and organic matter content = 1.7%, the commutative Cd release probability is 0.78.

[0050] In actual operation, if the weather forecast indicates that the pH will drop from 7.2 to 6.0 in the next 24 hours (rate of change = -1.2), after inputting this parameter, the belief propagation algorithm infers that the probability of exchangeable Cd release increases to 0.83. Combining the toxicity weighting factor Cd = 1.5, the weighted risk contribution is calculated to be 0.83 × 1.5 = 1.245. This is then weighted and summed with other forms and normalized to the [0,1] interval, yielding a comprehensive risk value of 0.71 for this sampling point. Through inverse distance weighted interpolation, a risk distribution map of the entire lake is generated, clearly showing high-risk patches in the northwest lake area.

[0051] Step Six: Deploy in-situ release monitoring devices in the aforementioned lake area, automatically collecting overlying water samples every 24 hours for 30 consecutive days. Actual measured data showed that the cumulative Cd release from day 5 to 7 was 0.26 μg / g, while the Bayesian model simulation value for the same period was 0.34 μg / g. The calculated RMSE was 0.18 > 0.15, triggering parameter correction. The Levenberg-Marquardt algorithm was used to invert and adjust the parameters in the conditional probability table of "pH change rate → probability of exchangeable Cd release". After three iterations, the RMSE decreased to 0.09, and the optimized CPT corrected the release probability under the scenario of rapid pH decrease from 0.83 to 0.76. The updated model output more closely resembles the actual measured trend, thus improving the reliability of the early warning.

[0052] Step 7: Because the optimized risk value of 0.71 exceeds the safety threshold of 0.65, the sampling frequency for this area is increased from monthly to weekly, and a LoRa IoT sensor array is deployed. The PH-801 sensor transmits real-time data at pH=6.1, the DO-2000 displays DO=4.8 mg / L, and the Cd²⁺ ion-selective electrode detects a concentration of 8.7 μg / L. The 24-hour sliding window mechanism triggers a model rerun, outputting a new risk value of 0.68; the updated value the next day is 0.70. Having exceeded the limit twice consecutively, the system immediately pushes an alarm to the management platform via the MQTT protocol, including coordinates (30.25°N, 120.15°E), the excess morphology (exchangeable Cd), the risk value (0.70), and recommended measures (addition of hydroxyapatite). Simultaneously, the control center activated the dosing pump to spray 5 g / m² of phosphate passivating agent onto the sediment surface. Within 72 hours, the Cd concentration in the overlying water decreased to 3.2 μg / L, thus verifying the effectiveness of the closed-loop response.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic assessment and early warning of heavy metal pollution risk of river and lake sediment, characterized in that, Includes the following steps: Data from sediment samples of the target water body were collected to obtain information on the distribution of heavy metal speciation. An initial dataset was constructed by combining sediment particle size, organic matter content, and water body ecological condition data. The water body ecological condition data included pH, temperature, and oxygen content. The distribution information of heavy metal speciation in the initial dataset is preprocessed, and a classification method is used to distinguish the chemical state change characteristics of different speciations to obtain the classified speciation dataset. Based on the classified morphological dataset, environmental change rate data is obtained. By simulating the dynamic fluctuations of aquatic ecological conditions, the stability of each morphology under different conditions is analyzed, and the trend of stability change is determined. Data on the influence of sediment particle size and organic matter content on the adsorption and release rate of heavy metals are extracted from the stability change trend. It is then determined whether the adsorption and release rate exceeds the preset threshold. If it does, it is marked as a high-risk form, and a risk form identifier set is generated. For the risk form identifier set, combined with environmental change rate and chemical state change data, a release risk value assessment model is constructed to calculate the potential release probability of each form under simulated environmental dynamic fluctuation conditions and determine the release risk value distribution. Based on the distribution of release risk values, prediction deviation data is obtained. By comparing the difference between the release amount under simulated conditions and the actual monitoring data, the parameters of the release risk value assessment model are adjusted to obtain the optimized risk prediction results. Using the optimized risk prediction results, a dynamic monitoring plan is generated for high-risk patterns, continuously updating the pattern stability and risk value data. If the risk value exceeds the safe range, an early warning mechanism is triggered through automatic notification and emergency response measures.

2. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The heavy metal speciation information includes the content ratios of exchangeable, carbonate-bound, iron-manganese oxide-bound, organic-bound, and residual forms; the classified speciation dataset is divided into speciation categories and labeled with their reactivity levels based on chemical extraction methods. The stability change trend includes the transformation path map of each form under pH gradient, dissolved oxygen fluctuation and temperature change; the risk form identifier set includes form type, corresponding sediment physicochemical parameters and critical release rate threshold; the release risk value assessment model adopts a Bayesian network structure to integrate multivariate inputs; the prediction deviation data quantifies the degree of deviation between the model output and the measured release amount through the root mean square error index; the dynamic monitoring scheme includes sampling frequency adaptive adjustment rules, sensor deployment location optimization strategy and risk threshold dynamic correction mechanism; the early warning mechanism pushes real-time alarms to the environmental management platform through message queue telemetry transmission protocol and triggers the response program of water quality control equipment.

3. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The specific steps for collecting sediment samples from the target water body, obtaining information on the distribution of heavy metal speciation, and constructing an initial dataset by combining sediment particle size, organic matter content, and water body ecological conditions data are as follows: Grid sampling points were set up in the target rivers and lakes, and sediment samples from a depth of 0–20 cm were obtained using columnar samplers. Multi-parameter water quality probes were deployed simultaneously to record the pH, dissolved oxygen concentration and water temperature at the sampling time, forming a spatiotemporally matched original observation record. After freeze-drying, the sediment samples were passed through a 100-mesh sieve. Five heavy metal species were separated by BCR continuous extraction. The concentrations of cadmium, lead, copper, zinc and chromium in each species were determined by inductively coupled plasma mass spectrometry, and a species distribution matrix was generated. Laser particle size analysis was performed on sediment samples from the same batch to obtain particle size distribution curves. The organic carbon content was determined by potassium dichromate oxidation-external heating method and converted into an organic matter content index. The morphological distribution matrix, particle size data, organic matter content, and water ecological parameters are structurally correlated according to the sampling point number to construct an initial dataset containing spatial coordinates, timestamps, and multidimensional attribute fields.

4. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The specific steps for preprocessing the distribution information of heavy metal speciation in the initial dataset and using a classification method to distinguish the chemical state change characteristics of different speciations to obtain the classified speciation dataset are as follows: The concentrations of each morphology in the initial dataset were normalized to eliminate dimensional differences, and principal component analysis was used to extract the main variation directions of morphology combinations. Based on principal component scores, the K-means clustering algorithm was used to divide heavy metal species into high-activity, medium-activity, and inert groups, and each group was assigned a stability label. A fuzzy C-means algorithm is introduced to assign membership degrees to boundary samples, avoiding information loss caused by hard partitioning and generating morphological classification results with fuzzy membership degrees. The classification results are bound to the physicochemical parameters of the sediment to form a classified morphological dataset containing morphological categories, membership coefficients, and corresponding environmental factors.

5. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The steps for obtaining environmental change rate data based on the classified morphological dataset, analyzing the stability of each morphology under different conditions by simulating the dynamic fluctuations of aquatic ecological conditions, and determining the trend of stability change are as follows: An environmental disturbance simulation chamber was constructed, and a three-factor orthogonal experimental matrix of pH, dissolved oxygen, and temperature was set. The classified sediment samples were placed in the chamber for dynamic exposure. Every N1 hours, samples of the overlying water were collected to determine the concentration of dissolved heavy metals, calculate the release flux per unit time, and generate a time-series release curve. Slope analysis was performed on the release curves to identify the inflection points of the release rates of each form under different environmental combinations, and a mapping table between the environmental change rate and the release rate was established. Grey relational analysis was used to quantify the influence weights of various environmental factors on the stability of different morphologies, and a heat map of the stability change trend was drawn.

6. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The steps of extracting data on the impact of sediment particle size and organic matter content on the adsorption and release rate of heavy metals from the stability change trend, determining whether the adsorption and release rate exceeds a preset threshold, and marking it as a high-risk form if it does, and generating a risk form identifier set are as follows: Based on the heat map of stability change trend, the experimental group with a release rate greater than the first release rate threshold was selected, and the corresponding median particle size D50 and organic matter content of the sediment were extracted. A multivariate nonlinear regression model was established with D50 and organic matter content as independent variables and release rate as dependent variable to fit the adsorption-desorption kinetic response surface. A second release rate threshold is set. When the model's predicted value exceeds this threshold, the corresponding morphology is marked as high risk. The type of high-risk morphology, its active group, critical D50 range, organic matter content range, and threshold parameters are encapsulated into structured entries to generate a risk morphology identifier set.

7. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The specific steps for constructing a release risk value assessment model based on the risk morphology identifier set, combined with environmental change rate and chemical state change data, to calculate the potential release probability of each morphology under simulated dynamic environmental fluctuation conditions, and to determine the release risk value distribution, are as follows: Using Bayesian networks as a framework, nodes are defined to include environmental change rate, sediment physicochemical parameters, and morphological categories, while edges represent causal dependencies. Conditional probability tables are trained using historical experimental data to determine the release probability distribution of child nodes under each combination of parent nodes; Input real-time or predicted aquatic ecological parameters and infer the release probability of each risk form through belief propagation algorithm; The release probability is multiplied by the morphological toxicity weight, and the weighted sum is obtained to obtain the comprehensive release risk value, generating a spatialized risk value distribution map.

8. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The steps of obtaining prediction deviation data based on the release risk value distribution, adjusting the parameters of the release risk value assessment model by comparing the difference between the release amount under simulated conditions and the actual monitoring data, and obtaining the optimized risk prediction result are as follows: In-situ release monitoring devices were deployed in the target water area to record the changes in heavy metal concentration in the overlying water for 30 consecutive days, and the measured cumulative release was calculated. Run the release risk value assessment model, input the environmental parameters for the same period, and output the simulated release amount; Calculate the root mean square error between the simulated value and the measured value. If the root mean square error is greater than the first error threshold, the parameter correction process is triggered. The Levenberg-Marquardt algorithm is used to invert and adjust the conditional probability parameters in the Bayesian network until the root mean square error is less than or equal to the second error threshold. Then, the optimized risk prediction result is output. The first error threshold is greater than the second error threshold.

9. The method for dynamic risk assessment and early warning of heavy metal pollution in river and lake sediments according to claim 1, characterized in that, The steps for generating a dynamic monitoring scheme for high-risk patterns using the optimized risk prediction results, continuously updating pattern stability and risk value data, and triggering an early warning mechanism through automatic notification and emergency response measures if the risk value exceeds the safe range are as follows: Set a risk value safety threshold. When the output value of the optimized model exceeds the threshold, the sampling frequency of the corresponding area will be automatically increased to once a week. Deploy IoT water quality sensor arrays in high-risk areas to transmit real-time data on pH, dissolved oxygen, temperature, and heavy metal ion concentrations. Establish a risk value sliding window update mechanism to refresh the prediction results every N2 hours based on new data; If two consecutive update values ​​exceed the threshold, a structured alarm is sent to the environmental monitoring platform, and an emergency response procedure for aeration or the addition of passivating agents is initiated.

Images