A method and system for assessing heavy metal pollution of a gulf marine environment
By using big data analytics, combined with techniques such as sliding window smoothing and feature filtering, we have solved the problems of accurately analyzing the distribution of heavy metal speciation and tracing pollution mechanisms in the Gulf marine environment, achieving high-precision pollution assessment and visualization.
Patent Information
- Application Number
- CN202610869004.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-16
- Publication Date
- 2026-08-25
AI Technical Summary
Existing technologies are insufficient to accurately analyze the distribution of heavy metal speciation and trace pollution mechanisms in the Gulf marine environment. In particular, the assessment results deviate significantly from the actual pollution situation under multi-factor coupled perturbation, and it is difficult to efficiently aggregate and clean multi-source monitoring data.
Using big data analytics, this study employs techniques such as sliding window smoothing, feature filtering, long short-term memory network mapping, and support vector machine association. Combined with environmental factors and metal composition data, a heavy metal pollution assessment method is constructed, including modules for environmental factor processing, metal speciation analysis, toxicity and historical correlation, pollution mechanism assessment, and spatial mapping.
It improved the accuracy of heavy metal speciation analysis, enabled traceability of pollution mechanisms, and enhanced the spatial visibility and timeliness of assessment results.
Smart Images

Figure CN122635978A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of computer data processing technology, specifically a method and system for assessing heavy metal pollution in the marine environment of a bay. Background Technology
[0002] The marine environment of the bay is located at the intersection of land and sea, and its ecological status is of great significance to the coastal areas. Heavy metal pollution assessment is a key link in measuring the environmental safety of this region. Currently, most common assessment methods directly measure the total amount of heavy metals in the water and compare it with the limit. These methods only focus on the apparent concentration and fail to reflect the impact of various environmental factors such as salinity, temperature, pH, and dissolved oxygen in the bay water on the actual form and toxicity of pollutants. They are prone to overlooking actual pollution conditions such as damage to benthic organisms even when the total amount meets the limit.
[0003] Furthermore, the aforementioned environmental factors in the bay waters are not independent of each other, but rather undergo continuous coupled changes with the seasons and tides. For example, rising water temperature causes a decrease in dissolved oxygen, which in turn alters the pH, thereby changing the form of heavy metals in the water, causing them to transform between insoluble and bioavailable forms. Because the relationship between environmental factors and the distribution of heavy metal forms is complex and nonlinear, and because instantaneous physical disturbances in areas such as thermoclines generate significant noise, conventional treatment methods struggle to accurately characterize this coupling mechanism, leading to substantial discrepancies between assessment results and the actual pollution situation.
[0004] On the other hand, bay monitoring often relies on a large number of sensor nodes deployed in different waters to collect data continuously over a long period of time. The resulting multi-source monitoring data, such as water temperature, dissolved oxygen, pH, and spectrum, is massive in scale, heterogeneous in origin, and has different sampling frequencies. It includes both current real-time data and long-term accumulated historical pollution archives. Conventional single-point comparison methods are unable to efficiently aggregate and clean such a large amount of data, and it is also difficult to retrieve evolution patterns that match the current pollution state from historical archives. This makes the assessment results lack the ability to trace the source and migration process of pollution.
[0005] Therefore, how to accurately analyze the distribution of heavy metal speciation and trace the pollution mechanism under multi-factor coupled perturbation using big data analysis has become a technical problem that urgently needs to be solved in the assessment of heavy metal pollution in the Gulf marine environment. Summary of the Invention
[0006] To address the above issues, this application provides a method and system for assessing heavy metal pollution in the marine environment of a bay, which solves the problem of how to accurately analyze the distribution of heavy metal speciation and trace the pollution mechanism under multi-factor coupled perturbation using big data analysis.
[0007] To achieve the above objectives, the technical solution adopted in this application is as follows: According to a first aspect of this application, a method for assessing heavy metal pollution in a bay marine environment is provided, comprising: S1. Water temperature, dissolved oxygen and pH were obtained from multiple sampling points at various depths in the bay. The parameter shift of the thermohaline was smoothed by a sliding window to obtain the second environmental factor data. S2. Perform feature filtering on the second environmental factor data and construct the interaction term between dissolved oxygen and pH to obtain the time series sequence of the first environmental factor with timestamps; S3. Obtain the total metal content of the bay water and analyze the total metal content according to its morphology to obtain the first metal component set; S4. Use a long short-term memory network to perform a nonlinear mapping between the time series of the first environmental factors and the set of the first metal components, and perform a delayed alignment on the hysteresis of the salinity coupling coefficient and the deposition flux to obtain the distribution characteristics of the first metal morphology. S5. When the ratio of complexed zinc to adsorbed mercury in the first metal morphology distribution characteristics exceeds the morphological stability threshold, toxicity classification is performed to obtain the first toxicity status with chronic exposure dose label. S6. Obtain historical pollution data corresponding to the first toxicity state, and perform time-series alignment of the evolution trajectory of benthic damage rate and food chain amplification factor to obtain a historical pollution sequence that matches the distribution characteristics of the first metal form. S7. Use support vector machine to perform correlation analysis between the first toxicity state and historical pollution sequences, extract the coupling relationship between the exceedance multiple of sediment layer and water column migration rate, and obtain the characteristics of the first pollution mechanism; S8. When the product of the enrichment index level and the potential ecological hazard value in the first pollution mechanism feature exceeds the safety threshold, a pollution degree assessment is performed and a regional risk correction term is fused to obtain the pollution assessment result; when it does not exceed the safety threshold, it is marked as a safety level and output. S9. Spatial interpolation is performed on the geospatial data corresponding to the pollution assessment results, and the bay coastline and water depth grid are overlaid to obtain a heavy metal pollution distribution map.
[0008] Preferably, S1 includes: The isolated forest algorithm was used to detect outliers in the acquired water temperature, dissolved oxygen, and pH levels to obtain the first environmental factor data. Calculate the parameter gradient between adjacent depths in the first environmental factor data, and determine the depth region where the parameter gradient is greater than a preset gradient threshold as a thermohaline. Extract the parameter offset within the thermohaline, determine the target sliding window period based on the comparison result between the parameter offset and the preset offset benchmark, and use the target sliding window period to perform weighted sliding window smoothing on the first environmental factor data within the thermohaline to obtain the second environmental factor data that removes instantaneous disturbances. When the smoothness index of the second environmental factor data is lower than the preset smoothing threshold, a second mean filter is applied.
[0009] Preferably, S2 includes: The random forest algorithm is used to calculate the feature importance of each factor in the second environmental factor data, and factors with feature importance higher than a preset importance threshold are retained; The dissolved oxygen and pH values in the retained factors are multiplied at the same collection time to obtain the synergistic value. The synergistic value is then bound to the corresponding timestamp and sorted in chronological order to obtain the synergistic change sequence. When the synergy value of consecutive time periods in the synergistic change sequence is greater than a preset synergy threshold, the time period is determined to be an environmental factor mutation and spliced with the synergistic change sequence to obtain a first environmental factor time series sequence with a timestamp.
[0010] Preferably, S3 includes: The total metal content was obtained by performing spectral analysis on water samples from the bay and then retrieving the data. The dissolved copper content, particulate lead content, suspended cadmium content, complexed zinc content, adsorbed mercury concentration, and colloidal arsenic ratio are separated from the total metal content to obtain the first metal component set; Concentration and proportion values are extracted from the first set of metal components to form a first feature matrix. A support vector machine is used to classify the pollution level of the first feature matrix, and the corresponding treatment priority is determined when the pollution level exceeds a preset threshold.
[0011] Preferably, S4 includes: The forgetting gate of a long short-term memory network is used to filter out noise in the time series of the first environmental factors whose influence weight is lower than a preset weight threshold. The time series of the first environmental factors is then nonlinearly mapped to the first set of metal components to obtain an initial mapping sequence. The time difference between the salinity coupling coefficient and the deposition interface flux in the initial mapping sequence is calculated to obtain the first hysteresis response time. When the first hysteresis response time is greater than a preset time threshold, the initial mapping sequence is shifted along the time axis to complete the delay alignment. Spatial deviation is calculated for the morphological distribution matrix after delayed alignment. When the maximum deviation exceeds the preset deviation threshold, numerical compensation is performed, and three-dimensional reconstruction is performed according to the depth coordinates of the deposition surface to obtain the first metal morphological distribution characteristics.
[0012] Preferably, in S5, the toxicity grading to obtain a first toxicity state including a chronic exposure dose label includes: The ratio of complexed zinc to adsorbed mercury was normalized, and the toxicity level was classified by the random forest algorithm. The obtained toxicity level was then concatenated with the chronic exposure dose identifier determined according to the exposure baseline to obtain the first toxicity state. S6 includes: using a dynamic time warping algorithm to time-align the evolution trajectory of the benthic organism damage rate and the food chain amplification factor to obtain the aligned evolution trajectory; The morphological matching degree between the alignment evolution trajectory and the first metal morphological distribution characteristics is calculated. When the morphological matching degree is lower than a preset matching threshold, the alignment evolution trajectory is linearly interpolated and compensated. The pollution amount corresponding to each time node after compensation is extracted to obtain the historical pollution sequence.
[0013] Preferably, S7 includes: The first toxicity state and the historical pollution sequence are concatenated into an input feature vector. A support vector machine with a radial basis function is used to map the input feature vector to a high-dimensional feature space. The coupling relationship between the exceedance multiple of the sediment layer and the water column migration rate is solved to obtain the first pollution mechanism feature. S8 includes: extracting the enrichment index level and potential ecological hazard value from the first pollution mechanism feature and multiplying them; when the product exceeds the safety threshold, using the Nemerow Integrated Pollution Index to assess the pollution level to obtain the initial pollution level; and fusing the initial pollution level with the regional risk level correction term determined according to the geographical location to obtain the first heavy metal pollution assessment result; when it does not exceed the safety threshold, marking it as a safe level and outputting it.
[0014] Preferably, S9 includes: The enrichment index level and potential ecological hazard value of each sampling point were spatially interpolated using the Kriging interpolation algorithm to obtain the contaminated continuous surface. When the coordinate system of the continuous polluted surface is inconsistent with that of the bay coastline and the water depth grid, the coordinate system of the bay coastline is transformed. The continuous polluted surface is masked and clipped using the bay coastline to remove land areas. The clipping result is then spatially superimposed with the water depth grid according to water depth to obtain a comprehensive pollution index. The comprehensive pollution index is classified into different levels according to the classification rules, and pixels that exceed the preset threshold are marked to obtain a distribution map of heavy metal pollution in the bay marine environment.
[0015] According to a second aspect of this application, a heavy metal pollution assessment system employing the aforementioned method for assessing heavy metal pollution in a bay marine environment is provided, comprising: The environmental factor processing module is configured to obtain water temperature, dissolved oxygen and pH from multiple sampling points in the bay water body, and smooth the temperature-salinity transition parameter offset by a sliding window to obtain the second environmental factor data. Then, feature filtering is performed on the data and an interaction term for dissolved oxygen and pH is constructed to obtain the first environmental factor time series with timestamps. The metal morphology analysis module is configured to obtain the total metal amount and analyze the total metal amount according to morphology to obtain the first metal component set. Then, a long short-term memory network is used to perform a nonlinear mapping between the time series of the first environmental factors and the first metal component set, and the lag alignment is performed by salinity coupling coefficient and deposition flux to obtain the distribution characteristics of the first metal morphology. The toxicity and history association module is configured to classify the toxicity of the first metal speciation distribution characteristics to obtain the first toxicity state when the ratio of complexed zinc to adsorbed mercury exceeds the threshold, and to obtain its historical pollution data, and to perform time-series alignment of benthic damage rate and food chain amplification factor trajectory to obtain the historical pollution sequence. The pollution mechanism assessment module is configured to use support vector machine to perform correlation analysis between the first toxicity state and historical pollution sequences to extract the coupling relationship between the exceedance multiple of sediment layer and water column migration rate, and to integrate regional risk correction term to assess the degree of pollution when the product of enrichment index level and potential ecological hazard value exceeds the threshold, thereby obtaining pollution assessment results; The spatial mapping module is configured to perform spatial interpolation on the geospatial data corresponding to the pollution assessment results and overlay shoreline and water depth grids to obtain a heavy metal pollution distribution map.
[0016] Preferably, the environmental factor processing module is further configured to: use the isolated forest algorithm to detect outliers in the acquired water temperature, dissolved oxygen and pH, calculate the parameter gradient between adjacent depths and determine the depth region where the parameter gradient is greater than a preset gradient threshold as a thermohaline layer, perform weighted sliding window smoothing on the environmental factors in the thermohaline layer according to a determined target sliding window period, and perform secondary mean filtering when the smoothness index is lower than a preset smoothing threshold to obtain the second environmental factor data.
[0017] The embodiments of this application have the following advantages: By applying a sliding window to smooth the parameter shift at the thermohaline layer, instantaneous physical disturbances caused by internal ocean waves are effectively eliminated, thus improving the stability of environmental factor data. By using nonlinear mapping of long short-term memory networks and delaying the alignment of the hysteretic response of salinity coupling coefficient and deposition interface flux, phase error in morphology mapping is eliminated, and the analytical accuracy of metal morphology distribution is improved. By using support vector machines to establish a high-dimensional correlation between toxicity status and historical pollution sequences, the coupling mechanism between the exceedance multiple of sedimentary layers and the water column migration rate was clearly characterized, thus achieving traceability of the pollution mechanism. By spatial interpolating and overlaying shoreline and water depth grids, discrete assessment results are transformed into continuous pollution distribution maps, improving the spatial visibility of the assessment results. Attached Figure Description
[0018] Figure 1 This is a flowchart of a method for assessing heavy metal pollution in the marine environment of a bay, as described in this application. Figure 2 This is a block diagram of the heavy metal pollution assessment system for the marine environment of the Gulf, as described in this application. Detailed Implementation
[0019] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.
[0020] The heavy metal pollution assessment method for the bay marine environment provided in this application can be implemented collaboratively by a network of sensor nodes deployed in the bay waters and a backend server. In an example system, the data acquisition layer consists of buoy sensor nodes deployed at different locations in the bay, used to collect multi-source data such as water temperature, dissolved oxygen, pH, and spectral data; the data processing layer receives the collected data and sequentially performs environmental factor denoising, metal speciation analysis, and pollution mechanism assessment; the distributed storage layer includes a time-series database and a historical pollution database, used to store intermediate data generated at each stage; the assessment application layer is responsible for generating pollution distribution maps and providing external queries. This application relies on big data analysis methods to aggregate, clean, and correlate massive amounts of multi-source monitoring data, thereby supporting the execution of the following steps.
[0021] Specifically, the big data analysis in this application is reflected in the following levels: First, at the data access level, water temperature, dissolved oxygen, pH, and spectral data from different buoy sensor nodes, at different depths, and with different sampling frequencies are uniformly aggregated into distributed storage and aligned and merged according to sampling point identifiers and timestamps to form a unified data view for subsequent processing; Second, at the data cleaning level, missing value imputation, outlier removal, and dimensional normalization are performed on the aggregated massive data to eliminate systematic biases between different sensor nodes; Third, at the correlation analysis level, the processed current monitoring data is correlated and retrieved across time periods with long-term accumulated pollution records in the historical pollution database to uncover pollution evolution patterns that match the current environmental state. All of the above processing levels can be executed in parallel within a distributed computing framework, thus maintaining the timeliness of the assessment even with large data volumes. Steps S1 to S9 below are based on this big data analysis.
[0022] See Figure 1The heavy metal pollution assessment method for the bay marine environment in this embodiment includes steps S1 to S9, with data being transferred step by step between each step, and the output of the previous step serving as the input of the next step, as detailed below.
[0023] Step S1: Obtain water temperature, dissolved oxygen, and pH from multiple depth sampling points in the bay water body, and smooth the parameter shift of the thermohaline layer using a sliding window to obtain the second environmental factor data.
[0024] In one implementation, the server receives water temperature, dissolved oxygen, and pH data collected every five meters from buoy sensor nodes within a 50-meter range from the water surface to underwater. An isolated forest algorithm is then used to remove significantly outlier samples, yielding the first environmental factor data. Specifically: After acquiring the first water temperature, first dissolved oxygen, and first pH from multiple depth sampling points, the server treats each depth's multidimensional environmental parameters as a single sample point to construct an isolated forest model. First, a subset of samples is randomly selected from all sample points as a sub-sample set. Cutting points are randomly generated within the value range of each feature dimension (water temperature, dissolved oxygen, pH), and the sub-sample set is recursively divided until each child node contains only one sample point or reaches the maximum tree depth, thus constructing an isolated tree. This process is repeated to generate multiple isolated trees, forming an isolated forest. For each sample point, its path length (the number of edges traversed from the root node to a leaf node) in each isolated tree is calculated, and the average of all tree path lengths is taken as the average depth. This average depth is normalized to the average path length of the total number of sample points in the isolated tree to obtain an anomaly score. The higher the anomaly score, the more likely the sample point is to be isolated (i.e., deviating from the normal distribution). The server identifies sample points with anomaly scores exceeding a preset threshold (e.g., 0.6) as significantly deviated abnormal sampling values and removes them. The remaining sample points constitute the first environmental factor data.
[0025] Subsequently, the parameter gradient between adjacent depths is calculated pairwise. When the temperature gradient of a certain depth interval exceeds a preset gradient threshold, that interval is identified as a thermohaline. Specifically: The server extracts the temperature values corresponding to each sampling depth from the first environmental factor data. Following the increasing depth order, it calculates the ratio of the temperature difference between adjacent depths to the depth interval to obtain the temperature gradient. For example, if the water temperature at 20 meters is 23.5℃ and at 25 meters is 22.7℃, with a 5-meter interval, the gradient is (23.5-22.7) / 5 = 0.16℃ / m. The calculated temperature gradient is compared with a preset gradient threshold (e.g., 0.15℃ / m). If the temperature gradient of a certain depth interval (e.g., 20-25 meters) exceeds this threshold, a thermohaline is identified in that interval, and its upper and lower boundary depths are recorded.
[0026] The preset gradient threshold is used to distinguish between the conventional water layer and the thermo-salinity layer. Its value is determined based on the formation conditions of the thermo-salinity layer in the bay water: the temperature change with depth at the thermo-salinity layer is significantly stronger than that at the conventional water layer. In engineering, the temperature difference between adjacent depths is usually taken as 0.5 degrees Celsius to 1 degree Celsius per meter as the criterion. In this embodiment, 0.8 degrees Celsius per meter is used. This value can reliably identify the thermo-salinity layer in the summer working conditions of most bays without misjudging the conventional fluctuations as the thermo-salinity layer.
[0027] Because the thermohaline zone is susceptible to instantaneous physical disturbances caused by internal ocean waves, this embodiment extracts the parameter offset within this interval and compares it with a preset offset benchmark. Based on this, the target sliding window period is determined, and then sliding window smoothing is performed on the first environmental factor data within this interval. The preset offset benchmark is adjusted based on historical undisturbed background monitoring data of the bay area: the fluctuation range of parameters such as water temperature, dissolved oxygen, and pH in the thermohaline zone under calm sea conditions is collected for at least one year, and the standard deviation or interquartile range of each parameter is taken as the benchmark value. For example, the temperature benchmark offset is ±0.3℃, and the dissolved oxygen benchmark offset is ±0.5mg / L. When the real-time parameter offset exceeds the benchmark value by a certain proportion (such as 1.5 times), it is determined that sliding window smoothing needs to be initiated. This benchmark value can be dynamically adjusted according to the season and tides, or it can be combined with three times the error of the sensor measurement accuracy as the lower limit.
[0028] The smoothing method for the sliding window is a Gaussian weighted moving average, which involves assigning Gaussian weights to each sampling point within the window based on its time distance from the center point and then calculating a weighted average. Specifically: ; in, Let be the smoothed output value at time t, x(t+k) be the sampled value within the window that deviates from the center by k sampling intervals, w(k) be the corresponding Gaussian weight, and m be the half-width of the window. The weights are set according to a Gaussian distribution, with the center point having the largest weight and decreasing towards both sides. In this embodiment, the window span is 60 minutes, containing seven consecutive sampling points, and the center point weight is 0.4, decreasing to 0.1 towards both sides according to a normal shape. The above window span and weight values are based on the following: the window span needs to be greater than the typical duration of the internal wave disturbance in order to smoothly filter it out, but it cannot be too large to avoid smoothing out the real water quality evolution trend. Sixty minutes is an empirical value that takes both into account; a larger weight for the center point can retain the real information of the center moment while suppressing noise.
[0029] The smoothness index is calculated as follows: In the output sequence after sliding window processing, the absolute value of the difference between two adjacent output values is calculated for each pair. The sum of these values is then divided by the number of differences to obtain the mean of the absolute values. The reciprocal of this mean is then taken. For example, in a sequence [5.8, 5.9, 6.0], the absolute values of adjacent differences are 0.1 and 0.1, the mean is 0.1, and the reciprocal is 10, which is the smoothness index. A higher index indicates a smoother sequence.
[0030] After smoothing, if the smoothness index of the resulting sequence is still lower than the preset smoothing threshold, a second mean filter is applied to obtain the second environmental factor data. The smoothness index is the reciprocal of the mean of the absolute values of the differences between adjacent output values within the window. The smoothing threshold is calibrated based on the sensor's measurement noise level, ensuring that the second filter is triggered only when residual jitter is significant.
[0031] Step S2: Perform multi-dimensional feature screening on the second environmental factor data and construct the interaction term between dissolved oxygen and pH to obtain the first environmental factor time series with timestamps.
[0032] In one implementation, the server uses a random forest algorithm to calculate the feature importance of each factor in the second environmental factor data, which includes multiple dimensions such as water temperature, turbidity, dissolved oxygen, and pH. For each node split in each decision tree, the Gini impurity before the split is calculated first, and then the weighted sum of the Gini impurities of the two child nodes after the split is calculated. The difference between the two is the amount of impurity reduction brought by that factor at that node. This reduction is multiplied by the proportion of the number of samples covered by that node to the total number of samples, and then added to the total contribution value of the corresponding factor. Every split node of all decision trees in the random forest is traversed, and the total cumulative Gini impurity reduction of each factor is counted. Finally, the cumulative values of all factors are normalized so that the sum of importance is 1. Based on a preset importance threshold (e.g., 5%), factors higher than this threshold are retained, and redundant features are removed.
[0033] The preset importance threshold is used to filter factors that contribute little to the evaluation. Its value is based on the distribution of importance scores: In this embodiment, the importance of core factors such as dissolved oxygen and pH is significantly higher than that of other factors. Therefore, the threshold is set at about five percent, which can remove noise factors while retaining core factors.
[0034] Subsequently, the dissolved oxygen and pH values were multiplied at the same acquisition time to obtain a synergistic value reflecting the nonlinear coupling strength between the two. The interaction term is in the form of a product because the combined effect of dissolved oxygen and pH on the heavy metal speciation is approximately a synergistic amplification relationship, and the product can amplify its contribution to the coupling strength when any factor deviates from the baseline.
[0035] Each collaborative value is bound to its corresponding timestamp and sorted chronologically to obtain a collaborative change sequence. When the collaborative value exceeds a preset collaborative threshold within a continuous time period, it is determined that a sudden change has occurred in the environmental state during that time period. The time period of the sudden change is then concatenated with the collaborative change sequence to obtain a first environmental factor time series sequence with timestamps. The preset collaborative threshold is adjusted based on the statistical distribution of historical collaborative values (e.g., taking the mean plus a certain number of standard deviations) to ensure that it is triggered only during significant changes.
[0036] Step S3: Obtain the total amount of metals in the bay water, and separate the morphological components of each heavy metal from the total amount of metals to obtain the first metal component set.
[0037] In one implementation, the server performs spectral detection on a bay water sample and inverts the spectral reflectance into the total metal content. The inversion is calculated by first establishing a quantitative relationship model between spectral reflectance and total metal content using a sample with known heavy metal concentrations. For example, partial least squares regression is used to fit the reflectance of multiple bands as independent variables and the total metal content as dependent variable, and then the measured reflectance is substituted into the model to obtain the total metal content.
[0038] Based on the total metal content, conventional speciation separation and detection methods in this field are used to separate the dissolved copper content, particulate lead content, suspended cadmium content, complexed zinc content, adsorbed mercury concentration, and colloidal arsenic ratio to obtain the first metal component set. For example, the concentration or ratio of each speciation component can be obtained through filtration, ultrafiltration, chemical extraction, or spectroscopic detection.
[0039] Those skilled in the art can select appropriate separation methods based on the available equipment, such as filtration separation, ultrafiltration / centrifugation separation, chemical extraction, ion exchange / solid phase extraction, anodic stripping voltammetry, and spectroscopic analysis. Further, the concentration values of each form of heavy metal and the proportion of colloidal arsenic (15.4%) are extracted from the first metal component set. These values are arranged in a fixed order, normalized, and form a six-dimensional first feature vector. The feature vectors of multiple sampling points are combined to form the first feature matrix. A trained support vector machine model (e.g., using a radial basis function kernel) is used to classify the pollution level of this feature matrix, and the model outputs pollution levels such as light, moderate, and heavy. If the pollution level obtained from the classification exceeds a preset level threshold, a treatment priority determination is triggered. This level threshold is set according to the limits of each heavy metal in standards such as "Marine Sediment Quality Standard" (GB 18668) and "Seawater Quality Standard" (GB 3097): for example, the sediment limit for copper is 35.0 mg / kg. If the measured value exceeds this limit by a certain multiple, the threshold level is correspondingly increased. Priority scores are calculated based on the degree of exceedance and the toxicity response coefficient. The higher the score, the higher the priority for treatment, thus determining the first priority for treatment.
[0040] Step S4: A long short-term memory network is used to perform a nonlinear mapping between the time series of the first environmental factors and the first set of metal components, and the hysteresis response of the salinity coupling coefficient and the deposition interface flux is delayed and aligned to obtain the distribution characteristics of the first metal morphology.
[0041] In one implementation, the server synchronously inputs a first environmental factor time series and a first metal component set into a long short-term memory network. This network uses a forget gate to filter out environmental noise whose influence weights are below a preset weight threshold, and performs a nonlinear mapping on the input sequence to obtain an initial mapped sequence. The weight threshold is used to suppress interference from weakly correlated noise on the mapping, and its value is determined based on the statistical distribution of the forget gate output; in this embodiment, it is approximately 10%.
[0042] The number of hidden layers and the number of neurons in each layer of the Long Short-Term Memory Network are adjustable network structure parameters. In this embodiment, three hidden layers with 128 neurons per layer are used. The values of the number of layers and the number of neurons are based on the following: too few layers make it difficult to fit the strong nonlinear relationship between environmental factors and metal morphology, while too many layers are prone to overfitting and increase computational overhead. Three layers with 128 neurons is an empirical value that balances fitting ability and generalization ability. Those skilled in the art can adjust it appropriately according to the data scale.
[0043] During feature extraction, the server extracts the salinity coupling coefficient and deposition interface flux from the initial mapping sequence and calculates their time difference to obtain the first hysteresis response time. The time difference is calculated by performing a cross-correlation operation on the two sequences and taking the time offset corresponding to the peak value of the cross-correlation function. When the first hysteresis response time exceeds a preset time threshold, the initial mapping sequence is shifted along the time axis to complete delay alignment, thereby eliminating the phase error between salinity changes and deposition interface flux response; the time threshold is determined based on the sampling interval and the tolerable phase error.
[0044] After delayed alignment, the spatial deviation of the morphology distribution matrix is calculated. Specifically, the morphology distribution matrix is obtained as follows: First, a Long Short-Term Memory (LSTM) network is used to perform a nonlinear mapping between the time series of the first environmental factors and the first set of metal components, resulting in an initial mapping sequence. This sequence contains the dynamic correlation characteristics between environmental factors (such as water temperature, dissolved oxygen, pH, salinity, etc.) and metal components (such as dissolved copper, particulate lead, etc.). After completing the hysteresis response detection of salinity coupling coefficient and deposition interface flux and time axis translation (delayed alignment), the system extracts the concentration or proportion data of different metal morphologies (such as exchangeable state, iron-manganese oxide bound state, etc.) at each sampling depth and sampling time from the delayed aligned sequence, and organizes them into a two-dimensional or three-dimensional array according to spatial location (sampling point) and morphology type, i.e., the first metal morphology distribution matrix. The rows of this matrix usually correspond to different sampling depths or spatial points, and the columns correspond to different metal chemical morphologies.
[0045] The metal morphology distribution matrix after delay alignment is subtracted element by element from the measured reference matrix at the same position to obtain the absolute deviation value of each morphology concentration at each sampling point; then the square of each deviation value is summed and the root mean square deviation is obtained; finally, the deviation results are arranged according to spatial coordinates to obtain the first morphology deviation matrix.
[0046] The sedimentary surface depth coordinates refer to the water depth (i.e., the vertical distance from sea level to the sediment surface) at each sampling point, such as 5 meters, 20 meters, and 50 meters underwater. These depth coordinates were determined in S1 when environmental factors (water temperature, dissolved oxygen, pH) were obtained from multiple depth sampling points and stored as location attributes in the data. In S4, before 3D reconstruction, the system obtains the sedimentary surface depth coordinates corresponding to the second metal morphology distribution matrix (i.e., calls the precise water depth values of each previously stored sampling point), and then maps the morphology data in the matrix to the corresponding depth layer in 3D space based on these coordinates, thereby constructing the first 3D morphology model (i.e., the first metal morphology distribution characteristics).
[0047] When the maximum deviation element in the first morphology deviation matrix exceeds a preset deviation threshold, numerical compensation is performed, followed by three-dimensional reconstruction based on the deposition surface depth coordinates to obtain the first metal morphology distribution characteristics. The deviation threshold is determined based on the reasonable variation range of metal morphology between adjacent spatial grids to identify and correct abnormally abrupt grid values.
[0048] Step S5: When the ratio of the complexed zinc content to the adsorbed mercury concentration in the first metal morphology distribution feature exceeds the morphological stability threshold, the first metal morphology distribution feature is classified into toxicity levels to obtain the first toxicity status with chronic exposure dose indicators.
[0049] In one implementation, the server obtains a ratio by dividing the content of complexed zinc by the concentration of adsorbed mercury. This ratio reflects the relative relationship between two different migratory forms; a higher ratio indicates that the morphological system deviates more from stability. When the ratio exceeds a preset morphological stability threshold, it is normalized and a random forest algorithm is used to classify the toxicity level. Specifically, the first morphological value is subtracted from the minimum value of this ratio in historical monitoring, and then divided by the difference between the maximum and minimum values, mapping it to the [0,1] interval to obtain the second morphological value. Subsequently, the second morphological value, along with other environmental features (such as temperature, pH, etc.), is input into a trained random forest classification model. This model contains multiple decision trees, each of which splits the features based on Gini impurity and outputs the probability of each toxicity level. The highest probability, either the mode or the average probability, of all tree predictions is taken as the final toxicity level, such as Level 1 (low toxicity), Level 2 (moderate toxicity), or Level 3 (high toxicity). Next, based on the preset exposure baseline, the corresponding chronic exposure dose label is determined, and the toxicity level and dose label are concatenated to obtain the first toxicity state. Specifically: the toxicity level output by the random forest (e.g., "Level III") and the chronic exposure dose label classified by the support vector machine (e.g., "0.015 mg / kg / d") are concatenated according to a preset format: for example, forming a key-value pair of "Level III|0.015", or encapsulating it into a structure containing level and dose fields to form a unified first toxicity state record, which facilitates subsequent historical pollution data retrieval and correlation analysis.
[0050] The morphological stability threshold is tuned based on the normal distribution characteristics of the ratio under current geological and hydrological conditions. For example, the quantile near the mean of the distribution is taken as the threshold so that it can distinguish between stable and unstable states. The exposure benchmark is determined based on the daily intake per unit body weight model under chronic exposure scenarios.
[0051] In this step, the ratio of complexed zinc to adsorbed mercury was selected as the trigger index for toxicity classification, mainly based on the following marine environmental chemistry and toxicology: First, zinc and mercury are typical heavy metal pollutants in the bay environment with significant differences in toxicity. Zinc often exists in a complexed form, reflecting the binding capacity of organic matter and bioavailability; while mercury has a strong tendency to be adsorbed by particles, and adsorbed mercury represents components in the sedimentary phase that are easily resuspended and have the potential risk of methylation.
[0052] The two morphologies are complementary, and changes in their ratio can comprehensively reflect the shift in the equilibrium of water-particle interface partitioning. Secondly, literature (such as studies in "Estuarine, Coastal and Shelf Science") indicates that under the coupled perturbation of dissolved oxygen and pH, the stability constant of zinc complexes and the partition coefficient of mercury adsorption exhibit a competitive response. Especially when a salinity stratum is developed, there is a significant time shift and ratio fluctuation between the complexation dissolution and release of zinc and the particle adsorption and desorption of mercury. When this ratio exceeds the morphological stability threshold, it often indicates a shift in the metal morphology in the water column from stable complexation dominance to adsorption particle release dominance, a precursor to increased toxicity.
[0053] Therefore, using the ratio of complexed zinc to adsorbed mercury as a sensitive indicator for triggering toxicity classification has clear mechanism orientation and early warning value, avoiding the one-sidedness of assessment based on a single metal or a single form.
[0054] Step S6: Obtain historical pollution data corresponding to the first toxicity state, and perform time-series alignment of the evolution trajectory of benthic organism damage rate and food chain amplification factor to obtain a historical pollution sequence that matches the distribution characteristics of the first metal morphology.
[0055] In one implementation, the server retrieves relevant historical pollution data from a historical pollution database based on the first toxicity state, and extracts the evolution trajectory of benthic organism damage rate and food chain amplification factor. Since the sampling frequencies of historical data and current data may be inconsistent, a dynamic time warping algorithm is used to align the two trajectories temporally. The calculation method is to construct a cumulative distance matrix between each time point of the two sequences and find the curved path that minimizes the cumulative distance. The time points are aligned along this path, thereby eliminating the time offset caused by the inconsistent sampling frequencies.
[0056] After alignment, the morphological matching degree between the alignment evolution trajectory and the current morphological distribution features is calculated. The morphological matching degree is calculated using cosine similarity. When the morphological matching degree is lower than the preset matching threshold, linear interpolation compensation is performed on the alignment evolution trajectory, and then the pollution amount corresponding to each time node after compensation is extracted to obtain the historical pollution sequence.
[0057] The matching threshold is determined based on the requirements for the accuracy of historical similar event retrieval. The higher the threshold, the more similar the historical fragments are to the current form.
[0058] Step S7: Use support vector machine to perform correlation analysis between the first toxicity state and the historical pollution sequence, extract the coupling relationship between the exceedance multiple of the sediment layer and the water column migration rate, and obtain the characteristics of the first pollution mechanism.
[0059] In one implementation, the association analysis between the first toxicity state and the historical contamination sequence specifically involves: the server concatenating the first toxicity state and the second historical contamination sequence into an input feature vector, and then performing analysis using a support vector machine with a configured radial basis function (RBF) kernel. The RBF kernel maps the original low-dimensional feature data nonlinearly to a higher-dimensional space, making the complex nonlinear associations in the low-dimensional space simpler and solvable in the high-dimensional space.
[0060] In high-dimensional space, Support Vector Machines (SVMs) solve an optimization problem to find the optimal regression hyperplane that minimizes the sum of the deviations of all sample points from the hyperplane. During the optimization process, only sample points that fall on or exceed the boundary become support vectors. These support vectors represent the most representative correlation pattern between sedimentary exceedance factor and water column migration rate: for example, when the exceedance factor is low, the migration rate increases slowly; while when the exceedance factor exceeds a certain critical value, the migration rate shows an accelerated upward trend or exhibits nonlinear characteristics such as saturation.
[0061] Support Vector Machines (SVMs) utilize these key support vectors to construct a regression function, directly quantifying the coupling relationship between the two. The system extracts the weight coefficient matrix from the regression function, flattens and normalizes it, generating the first pollution mechanism feature. This feature not only reflects the strength of the positive correlation between the exceedance multiple of the sediment layer and the water column migration rate, but also characterizes the nonlinear sensitive region in their interaction, thus providing a clear mechanistic basis for subsequent pollution level assessment.
[0062] The basis for using the radial basis function is that there is a nonlinear positive correlation between the excess of sediment layer and the water column migration rate, and the radial basis function can characterize this nonlinear relationship without explicitly constructing a high-dimensional mapping; the dimension of the high-dimensional feature space is an adjustable parameter that needs to be matched with the sample size to avoid underfitting or overfitting.
[0063] Step S8: When the product of the enrichment index level and the potential ecological hazard value in the first pollution mechanism feature exceeds the safety threshold, the pollution degree of the first pollution mechanism feature is assessed, and a regional risk level correction term is integrated to obtain the first heavy metal pollution assessment result. When it does not exceed the safety threshold, it is marked as a safe level and output.
[0064] In one implementation, the server extracts the enrichment index level and potential ecological hazard value from the first pollution mechanism feature and multiplies them. This product reflects both the degree of pollution accumulation and the degree of ecological hazard, and is therefore compared with a safety threshold to determine whether further assessment is needed. The safety threshold is determined based on an acceptable level of ecological risk. When the product exceeds the safety threshold, the Nemerow Comprehensive Pollution Index is used to assess the degree of pollution. Its calculation method comprehensively considers the average and maximum values of individual pollution indices at each sampling point within the region, specifically: ; Wherein, P is the Nemerow Comprehensive Pollution Index, P(avg) is the average value of each individual pollution index within the region, and P(max) is the maximum value of each individual pollution index. The Nemerow Index incorporates both the average and maximum values, reflecting both the overall pollution level and highlighting the most severely polluted areas, thus making it suitable for characterizing the overall pollution level of a region and obtaining the initial pollution level. Subsequently, the initial pollution level is multiplied and fused with a regional risk level correction term determined according to the geographical location to obtain the first heavy metal pollution assessment result.
[0065] The regional risk level correction item is determined based on geospatial attributes such as the distance between the location and sensitive targets (such as water sources). The closer the location is to the sensitive target, the larger the correction item value is, so as to reflect the risk difference corresponding to the same level of pollution in different regions. The correspondence between various regions and correction item values is pre-set in a configurable manner in the system and can be adjusted according to management requirements.
[0066] Step S9: Based on the geospatial data corresponding to the first heavy metal pollution assessment results, spatial interpolation is performed on the enrichment index level and potential ecological hazard value of each sampling point, and the bay coastline and water depth grid are superimposed to obtain a distribution map of heavy metal pollution in the bay marine environment.
[0067] In one implementation, the server reads the latitude and longitude coordinates of each sampling point, along with its enrichment index level and potential ecological hazard value. It then uses the Kriging interpolation algorithm to interpolate the discrete point data into a continuous polluted surface. The Kriging interpolation fits the spatial autocorrelation using a spherical semivariogram model and sets the search radius and output pixel size. The search radius is determined based on the spatial density of the sampling points; too small a radius will result in some areas having no effective samples participating in the interpolation, while too large a radius will introduce weakly correlated samples. In this embodiment, a radius of 5 kilometers is used. The output pixel size is determined based on the required spatial resolution; in this embodiment, 30 meters is used.
[0068] When the coordinate systems of the continuous polluted surface, the bay coastline, and the water depth grid are inconsistent, the coordinate system of the bay coastline is transformed. Then, the continuous polluted surface is masked and clipped using the bay coastline to remove land areas. The clipping result is then superimposed with the water depth grid according to water depth weights to obtain a comprehensive pollution index. The pollution weight of shallow water areas is greater than that of deep water areas to reflect the more direct ecological impact of pollution in shallow water areas. Finally, the comprehensive pollution index is classified into levels according to the classification rules, and pixels exceeding the preset threshold are marked to obtain a distribution map of heavy metal pollution in the bay marine environment.
[0069] The weights are pre-configured based on the ecological sensitivity and pollution exposure risk of shallow and deep water areas. The tuning method can employ the analytic hierarchy process (AHP), inviting marine ecology experts to conduct pairwise comparisons and scoring of water depth classifications (e.g., 0–10m for shallow water, 10–30m for transitional water, and >30m for deep water), calculating the weight coefficient for each depth. Alternatively, based on the statistical law that benthic abundance decreases with depth, a weight of 0.6 can be set for shallow water and 0.4 for deep water. The weight values support dynamic adjustment to ensure consistency with the ecological reality of the assessment area.
[0070] In a specific example of this application, under spring conditions in an estuarine bay, the intensity of the thermohaline layer is relatively weak. Therefore, in step S1, the gradient threshold for determining the thermohaline layer is appropriately lowered, and the sliding window span is shortened to approximately thirty minutes with five sampling points to more sensitively track faster water quality evolution. In step S4, the Long Short-Term Memory network can adopt a smaller structure with two hidden layers and sixty-four neurons per layer to adapt to the relatively small data scale in this scenario. The remaining steps are the same as in Example 1. This embodiment shows that the processing chain of this application can be adapted to different bay environments by adjusting a few parameters, while the data transfer relationship between its preceding and following steps remains unchanged.
[0071] In another example of this application, the environmental factor denoising and metal morphology analysis in steps S1 to S4 are performed on edge computing nodes deployed near the buoy. Only the first metal morphology distribution features are uploaded to the backend server to complete steps S5 to S9. This reduces the amount of data transmitted to the backend and lowers transmission latency. In this deployment mode, after the edge computing nodes complete the sliding window smoothing in step S1 and the delay alignment in step S4 locally, they only need to upload the first metal morphology distribution features with a dimension significantly lower than the original sampling data. This allows the assessment process to continue running even in marine environments with limited network bandwidth or unstable transmission. After receiving the first metal morphology distribution features from multiple edge computing nodes, the backend server then uniformly executes steps S5 to S9 for toxicity classification, historical correlation, mechanism assessment, and spatial mapping to fully utilize the historical pollution database and global geospatial data. Simultaneously, the output pixel size of the Kriging interpolation in step S9 can be adjusted to approximately ten meters to obtain a more detailed distribution map, and the search radius is adjusted accordingly. This embodiment demonstrates that the steps of this application can be flexibly distributed between the edge and the backend without affecting the overall assessment logic and the data transmission relationship between preceding and subsequent steps.
[0072] This application also provides a system for assessing heavy metal pollution in the marine environment of a bay. See [link to relevant documentation]. Figure 2 The system includes an environmental factor processing module, a metal morphology analysis module, a toxicity and historical correlation module, a pollution mechanism assessment module, and a spatial mapping module.
[0073] The environmental factor processing module is configured to execute the above steps S1 and S2, that is, to complete the temperature-salinity gradient denoising and interaction term construction, and output the first environmental factor time series sequence. The metal morphology analysis module is configured to execute steps S3 and S4, namely, to complete the separation of metal morphology components and the mapping and delay alignment of the long short-term memory network, and output the first metal morphology distribution features. The toxicity and history association module is configured to execute steps S5 and S6, that is, to complete the toxicity classification and historical pollution sequence time alignment. The pollution mechanism assessment module is configured to execute steps S7 and S8, namely, to complete the mechanism coupling extraction and pollution degree assessment. The spatial mapping module is configured to execute step S9, which is to complete spatial interpolation and raster overlay and output a pollution distribution map.
[0074] Data is transferred hierarchically between modules, with the output of one module serving as the input of the next. The above module division is merely exemplary; those skilled in the art can merge or split the modules as needed without affecting their functionality.
[0075] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.
Claims
1. A method for assessing heavy metal pollution in a bay marine environment, characterized in that, include: S1. Water temperature, dissolved oxygen and pH were obtained from multiple sampling points at various depths in the bay. The parameter shift of the thermohaline was smoothed by a sliding window to obtain the second environmental factor data. S2. Perform feature filtering on the second environmental factor data and construct the interaction term between dissolved oxygen and pH to obtain the time series sequence of the first environmental factor with timestamps; S3. Obtain the total metal content of the bay water and analyze the total metal content according to its morphology to obtain the first metal component set; S4. Use a long short-term memory network to perform a nonlinear mapping between the time series of the first environmental factors and the set of the first metal components, and perform a delayed alignment on the hysteresis of the salinity coupling coefficient and the deposition flux to obtain the distribution characteristics of the first metal morphology. S5. When the ratio of complexed zinc to adsorbed mercury in the first metal morphology distribution characteristics exceeds the morphological stability threshold, toxicity classification is performed to obtain the first toxicity status with chronic exposure dose label. S6. Obtain historical pollution data corresponding to the first toxicity state, and perform time-series alignment of the evolution trajectory of benthic damage rate and food chain amplification factor to obtain a historical pollution sequence that matches the distribution characteristics of the first metal form. S7. Use support vector machine to perform correlation analysis between the first toxicity state and historical pollution sequences, extract the coupling relationship between the exceedance multiple of sediment layer and water column migration rate, and calculate the enrichment index level and potential ecological hazard value to obtain the characteristics of the first pollution mechanism. S8. When the product of the enrichment index level and the potential ecological hazard value exceeds the safety threshold, a pollution degree assessment is performed and a regional risk correction term is integrated to obtain the pollution assessment result; when it does not exceed the safety threshold, it is marked as a safety level and output. S9. Spatial interpolation is performed on the geospatial data corresponding to the pollution assessment results, and the bay coastline and water depth grid are overlaid to obtain a heavy metal pollution distribution map.
2. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S1 includes: The isolated forest algorithm was used to detect outliers in the acquired water temperature, dissolved oxygen, and pH levels to obtain the first environmental factor data. Calculate the parameter gradient between adjacent depths in the first environmental factor data, and determine the depth region where the parameter gradient is greater than a preset gradient threshold as a thermohaline. Extract the parameter offset within the thermohaline, determine the target sliding window period based on the comparison result between the parameter offset and the preset offset benchmark, and use the target sliding window period to perform weighted sliding window smoothing on the first environmental factor data within the thermohaline to obtain the second environmental factor data that removes instantaneous disturbances. When the smoothness index of the second environmental factor data is lower than the preset smoothing threshold, the second environmental factor data is obtained by performing a second mean filter.
3. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S2 include: The random forest algorithm is used to calculate the feature importance of each factor in the second environmental factor data, and factors with feature importance higher than a preset importance threshold are retained; The dissolved oxygen and pH values in the retained factors are multiplied at the same collection time to obtain the synergistic value. The synergistic value is then bound to the corresponding timestamp and sorted in chronological order to obtain the synergistic change sequence. When the synergy value of consecutive time periods in the synergistic change sequence is greater than a preset synergy threshold, the time period is determined to be an environmental factor mutation and spliced with the synergistic change sequence to obtain a first environmental factor time series sequence with a timestamp.
4. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S3 include: The total metal content was obtained by performing spectral analysis on water samples from the bay and then retrieving the data. The dissolved copper content, particulate lead content, suspended cadmium content, complexed zinc content, adsorbed mercury concentration, and colloidal arsenic ratio are separated from the total metal content to obtain the first metal component set.
5. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S4 includes: The forgetting gate of a long short-term memory network is used to filter out noise in the time series of the first environmental factors whose influence weight is lower than a preset weight threshold. The time series of the first environmental factors is then nonlinearly mapped to the first set of metal components to obtain an initial mapping sequence. The time difference between the salinity coupling coefficient and the deposition interface flux in the initial mapping sequence is calculated to obtain the first hysteresis response time. When the first hysteresis response time is greater than a preset time threshold, the initial mapping sequence is shifted along the time axis to complete the delay alignment, and the morphological distribution is calculated. The spatial deviation of the morphological distribution matrix after delayed alignment is calculated. When the maximum spatial deviation is greater than the preset deviation threshold, numerical compensation is performed, and three-dimensional reconstruction is performed according to the depth coordinates of the deposition surface to obtain the first metal morphological distribution characteristics.
6. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S6 includes: using a dynamic time warping algorithm to time-align the evolution trajectory of the benthic organism damage rate and the food chain amplification factor to obtain the aligned evolution trajectory; The morphological matching degree between the alignment evolution trajectory and the first metal morphological distribution characteristics is calculated. When the morphological matching degree is lower than a preset matching threshold, the alignment evolution trajectory is linearly interpolated and compensated. The pollution amount corresponding to each time node after compensation is extracted to obtain the historical pollution sequence.
7. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S7 includes: The first toxicity state and the historical pollution sequence are concatenated into an input feature vector. A support vector machine with a radial basis function is used to map the input feature vector to a high-dimensional feature space. The coupling relationship between the exceedance multiple of the sediment layer and the water column migration rate is solved to obtain the first pollution mechanism feature. The enrichment index level is obtained by taking the logarithm to base 2 of the ratio of heavy metal concentration in sediment to regional geochemical background value, and the potential ecological hazard value is the sum of the ratio of the concentration of each heavy metal to its sediment quality baseline multiplied by the toxicity response coefficient of that heavy metal.
8. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S8 includes: extracting the enrichment index level and potential ecological hazard value from the first pollution mechanism characteristics and multiplying them; when the product exceeds the safety threshold, using the Nemerow Integrated Pollution Index to assess the pollution level to obtain the initial pollution level; and merging the initial pollution level with the regional risk level correction term determined according to the geographical location to obtain the first heavy metal pollution assessment result.
9. The method for assessing heavy metal pollution in the marine environment of a bay according to claim 1, characterized in that, S9 includes: The enrichment index level and potential ecological hazard value of each sampling point were spatially interpolated using the Kriging interpolation algorithm to obtain the contaminated continuous surface. When the coordinate system of the continuous polluted surface is inconsistent with that of the bay coastline and the water depth grid, the coordinate system of the bay coastline is transformed. The continuous polluted surface is masked and clipped using the bay coastline to remove land areas. The clipping result is then spatially superimposed with the water depth grid according to water depth to obtain a comprehensive pollution index. The comprehensive pollution index is classified into different levels according to the classification rules, and pixels that exceed the preset threshold are marked to obtain a distribution map of heavy metal pollution in the bay marine environment.
10. A heavy metal pollution assessment system employing the heavy metal pollution assessment method for the bay marine environment according to any one of claims 1 to 9, characterized in that, include: The environmental factor processing module is configured to obtain water temperature, dissolved oxygen and pH from multiple sampling points in the bay water body, and smooth the temperature-salinity transition parameter offset by a sliding window to obtain the second environmental factor data. Then, feature filtering is performed on the data and an interaction term for dissolved oxygen and pH is constructed to obtain the first environmental factor time series with timestamps. The metal morphology analysis module is configured to obtain the total metal amount and analyze the total metal amount by morphology to obtain a first metal component set. It then uses a long short-term memory network to perform a nonlinear mapping between the time series of the first environmental factors and the first metal component set, and performs delayed alignment by using the salinity coupling coefficient and the lag of the deposition flux to obtain the distribution characteristics of the first metal morphology. Concentration values and proportion values are extracted from the first metal component set to form a first feature matrix. A support vector machine is used to classify the pollution level of the first feature matrix, and the corresponding treatment priority is determined when the pollution level exceeds a preset level threshold. The toxicity and history association module is configured to classify the toxicity of the first metal speciation distribution characteristics to obtain the first toxicity state when the ratio of complexed zinc to adsorbed mercury exceeds the threshold, and to obtain its historical pollution data, and to perform time-series alignment of benthic damage rate and food chain amplification factor trajectory to obtain the historical pollution sequence. The pollution mechanism assessment module is configured to use support vector machine to perform correlation analysis between the first toxicity state and historical pollution sequences to extract the coupling relationship between the exceedance multiple of sediment layer and water column migration rate, and to integrate regional risk correction term to assess the degree of pollution when the product of enrichment index level and potential ecological hazard value exceeds the threshold, thereby obtaining pollution assessment results; The spatial mapping module is configured to perform spatial interpolation on the geospatial data corresponding to the pollution assessment results and overlay shoreline and water depth grids to obtain a heavy metal pollution distribution map.
