An isotopic-based method for tracing surface water pollution

Abstract

The application discloses a kind of surface water pollution tracing methods based on isotope, it is related to environmental monitoring technical field, including, to isotope residual distribution diagram is carried out spatial clustering analysis, obtains time evolution characteristic table, time evolution characteristic table is executed space-time coding, obtains abnormal event sequence;Analysis time continuity and spatial variation of abnormal event sequence, output pollution propagation characteristic table, pollution propagation characteristic table is carried out space-time similarity matching and node association, generates pollution propagation association matrix;Deduce the link direction of pollution propagation association matrix, generate pollution propagation path set, pollution propagation path set is executed reverse tracking and regional compilation, output surface water pollution tracing report.The application is realized by constructing the isotope background model based on space-time environmental characteristic set and analyzing pollution propagation characteristics, chain reconstruction and dynamic tracking in time and space dimensions of pollution event.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a method for tracing the source of surface water pollution based on isotopes. Background Technology

[0002] Source tracing of surface water pollution is a crucial link in ecological environment monitoring and governance. Its core objective is to accurately identify pollution sources and trace pollution transmission pathways through technological means, providing a scientific basis for pollution prevention and control. Traditional pollution source tracing methods mainly include physical tracing, chemical fingerprinting, and microbial labeling. In recent years, stable isotope technology has gradually become a research hotspot due to its inherent advantage of "atomic-level tracing." Hydrogen and oxygen stable isotopes can trace water cycle processes, carbon and nitrogen stable isotopes can effectively identify the sources of organic pollutants, and radioactive isotopes are suitable for analyzing groundwater age and migration pathways. With the improvement of isotope detection equipment accuracy and the explosive growth of data volume, isotope-based pollution source tracing methods are gradually shifting from theoretical exploration to practical application.

[0003] However, existing technologies still have some shortcomings. Most methods rely on static isotope background values ​​to establish discrimination criteria, without fully considering the impact of temporal fluctuations in environmental factors such as meteorology and hydrology on isotope fractionation. This makes it difficult to capture the dynamic evolution of pollution events, resulting in source tracing results lagging behind actual pollution spread. Furthermore, the spatiotemporal correlation mining of multi-dimensional environmental data is insufficient, and isotope anomaly areas are analyzed in isolation, which fails to reveal the spatiotemporal coupling mechanism of pollution transmission and makes it difficult to accurately locate the spatial range and propagation path of pollution source areas. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an isotope-based method for tracing the source of surface water pollution to solve the problems of delayed source tracing and the inability to accurately reveal the spatiotemporal coupling mechanism of pollution propagation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for tracing the source of surface water pollution based on isotopes, which includes collecting hydrological and meteorological data and measured isotope data and performing preprocessing, and extracting the spatiotemporal environmental feature set of the preprocessed hydrological and meteorological data.

[0008] The spatiotemporal environmental feature set and preprocessed measured isotope data are input into the isotope background model. The isotope background model performs time-series dependency modeling and spatial correlation inference to generate an isotope residual distribution map.

[0009] Spatial clustering analysis was performed on the isotope residual distribution map to obtain a time evolution feature table. Spatiotemporal coding was performed on the time evolution feature table to obtain the sequence of anomalous events.

[0010] Analyze the temporal continuity and spatial variation of abnormal event sequences, output a pollution propagation feature table, perform spatiotemporal similarity matching and node association on the pollution propagation feature table, and generate a pollution propagation association matrix;

[0011] The link directions of the pollution propagation correlation matrix are derived, a set of pollution propagation paths is generated, reverse tracing and regional compilation are performed on the pollution propagation path set, and a surface water pollution source tracing report is output.

[0012] As a preferred embodiment of the isotope-based surface water pollution tracing method of the present invention, the hydrological and meteorological data includes meteorological observation data, hydrological process data, geospatial data, water quality physicochemical parameter data, and isotope observation data.

[0013] The preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

[0014] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the specific steps for extracting the spatiotemporal environmental feature set of the preprocessed hydrological and meteorological data are as follows:

[0015] By analyzing the fluctuation trends of preprocessed meteorological observation data, a meteorological feature set is generated;

[0016] Hydrological parameters are extracted from the preprocessed hydrological process data, and hydrological response analysis is performed to generate a hydrological feature set.

[0017] Spatial location encoding is performed on the preprocessed geospatial data to generate a spatial feature set;

[0018] Analyze the aggregated fluctuations of pretreated water quality physicochemical parameter data and output an environmental response feature set.

[0019] Baseline correction and migration analysis were performed on the preprocessed isotope observation data to obtain a time series feature set;

[0020] The meteorological feature set, hydrological feature set, spatial feature set, environmental response feature set and temporal feature set are structurally integrated to generate a spatiotemporal environmental feature set.

[0021] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the method involves inputting a spatiotemporal environmental feature set and preprocessed measured isotope data into an isotope background model. The isotope background model performs time-series dependency modeling and spatial correlation inference to generate an isotope residual distribution map. The specific steps are as follows:

[0022] The model parameters are initialized based on the spatiotemporal environmental feature set, and supervised training is performed using preprocessed measured isotope data to construct an isotope background model.

[0023] The spatiotemporal environmental feature set is input into the isotopic background model to analyze the temporal correlation and spatial dependence of the spatiotemporal environmental feature set and generate the expected isotopic value.

[0024] The preprocessed measured isotope data are compared and analyzed with the expected isotope values, and the isotope residual data are output.

[0025] Spatial clustering and pattern recognition are performed on isotopic residual data to generate isotopic residual distribution maps.

[0026] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the specific steps for performing spatial cluster analysis on the isotope residual distribution map to obtain a temporal evolution characteristic table are as follows.

[0027] Residual intensity aggregation and connected component partitioning are performed on the isotopic residual distribution map to obtain a set of isotopic anomaly regions.

[0028] Analyze the spatial correlation of isotopic anomaly regions and output a table of temporal evolution characteristics.

[0029] As a preferred embodiment of the isotope-based surface water pollution tracing method of the present invention, the specific steps for performing spatiotemporal coding on the time evolution feature table to obtain the abnormal event sequence are as follows:

[0030] Identify abnormal fluctuations in the time evolution feature table and perform time-series sorting to obtain an abnormal event window table;

[0031] Spatial clustering and event persistence analysis are performed on the abnormal event window table to generate an intensity feature table;

[0032] Analyze the spatiotemporal correlation of the intensity feature table and perform serialization processing to output the abnormal event sequence.

[0033] As a preferred embodiment of the isotope-based surface water pollution tracing method of the present invention, the specific steps for analyzing the temporal continuity and spatial variation of abnormal event sequences and outputting a pollution propagation characteristic table are as follows.

[0034] Analyze the temporal continuity of abnormal event sequences and generate an abnormal event feature table;

[0035] Identify the spatial clustering of abnormal event sequences and perform spatial variation trend analysis, outputting a spatial variation feature table;

[0036] The abnormal event feature table and the spatial change feature table are indexed and matched to obtain a spatiotemporal feature joint table. Spatiotemporal correlation analysis is performed on the spatiotemporal feature joint table to generate a pollution transmission feature table.

[0037] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the specific steps for generating a pollution propagation association matrix by performing spatiotemporal similarity matching and node association on the pollution propagation feature table are as follows.

[0038] Calculate the spatiotemporal similarity of the pollution transmission characteristic table, perform event matching based on the spatiotemporal similarity, and output the spatiotemporal matching result table;

[0039] Nearest neighbor search and spatiotemporal feature fusion are performed on the spatiotemporal matching result table to obtain the pollution transmission correlation matrix.

[0040] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the specific steps for deriving the link directions of the pollution propagation correlation matrix and generating a pollution propagation path set are as follows:

[0041] The pollution transmission correlation matrix is ​​processed by time-series weighting and node sorting to obtain a time-series link table.

[0042] Perform directional consistency filtering and spatial path reconstruction on the temporal link table to generate an initial propagation path set;

[0043] Verify the continuity of the initial propagation path set and perform topology optimization to output the pollution propagation path set.

[0044] As a preferred embodiment of the isotope-based surface water pollution source tracing method of the present invention, the specific steps for performing reverse tracing and regional compilation of the pollution propagation path set and outputting a surface water pollution source tracing report are as follows.

[0045] Perform reverse path analysis and node backtracking on the pollution transmission path set, and output the source tracing node set;

[0046] The source area feature table is generated by comparing the isotopic characteristics of the source node set with the pretreated water quality physicochemical parameter data and analyzing the correlation with environmental parameters.

[0047] Spatial clustering and regional boundary fitting are performed on the source area feature table to obtain a pollution source area distribution map;

[0048] The pollution source area distribution map is overlaid with the pollution propagation path set to generate a surface water pollution source tracing report.

[0049] The beneficial effects of this invention are as follows: By constructing an isotopic background model based on a spatiotemporal environmental feature set, time-series dependency modeling and spatial correlation inference are performed on multi-source meteorological, hydrological, and environmental data to obtain expected isotopic values ​​and generate isotopic residual distribution maps, effectively separating natural fluctuations and pollution disturbance signals, and improving the accuracy and stability of abnormal isotope identification; through the construction of abnormal event sequences and pollution propagation characteristic analysis, chain reconstruction and dynamic tracking of pollution events in the time and space dimensions are realized, revealing the pollution diffusion path and source area characteristic distribution, and improving the accuracy and interpretability of surface water pollution source tracing. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of an isotope-based method for tracing the source of surface water pollution.

[0052] Figure 2 The flowchart for generating the isotopic background model and isotopic residual distribution map.

[0053] Figure 3 A flowchart for generating a correlation matrix between anomaly event sequences and contamination propagation.

[0054] Figure 4 A flowchart for outputting a pollution transmission path reconstruction and surface water pollution source tracing report. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0058] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for tracing the source of surface water pollution based on isotopes, including the following steps:

[0059] S1. Collect hydrological and meteorological data and measured isotope data and preprocess them to extract the spatiotemporal environmental feature set of the preprocessed hydrological and meteorological data.

[0060] S1.1 Hydrological and meteorological data include meteorological observation data, hydrological process data, geospatial data, water quality physicochemical parameter data, and isotope observation data;

[0061] It should be noted that meteorological observation data such as rainfall, temperature, humidity, and evaporation are acquired in real time through automatic weather monitoring stations; hydrological process data are continuously recorded through flow monitoring sections, flow velocity meters, and water level sensors; geospatial data such as river network structure, terrain elevation, and watershed boundary information are extracted using remote sensing images; water quality physicochemical parameters such as conductivity, pH, dissolved oxygen, total nitrogen, total phosphorus, and chemical oxygen demand are acquired through online monitoring; and water samples are collected periodically, and isotope observation data are determined using an isotope mass spectrometer.

[0062] S1.2 Preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

[0063] It should be noted that data cleaning is performed on data from different monitoring devices and sampling periods to remove missing, duplicate, and invalid records; the cleaned data is then converted to a unified format to ensure consistency in timestamps, coordinates, and units; hash matching is used to deduplicate data and eliminate duplicate entries; the deduplicated data is then normalized to ensure comparability across all feature dimensions; and for detected outlier data points, outlier correction is performed using a combination of time series smoothing and spatial interpolation to restore continuity and physical plausibility while maintaining the original trend of the data.

[0064] S1.3. By analyzing the fluctuation trend of the preprocessed meteorological observation data, a meteorological feature set is generated.

[0065] It should be noted that the preprocessed meteorological observation data are sorted according to time index and smoothed by weighted moving average using the sliding window method to construct a meteorological observation sequence. The length of the sliding window is determined based on the meteorological monitoring sampling frequency and the short-term weather change cycle, such as 5 to 7 days. The sliding standard deviation and range of the meteorological observation sequence are used as quantitative standards for fluctuation amplitude. The larger the values ​​of the sliding standard deviation and range, the more drastic the meteorological changes, and the smaller the values ​​of the sliding standard deviation and range, the more stable the climate. The coefficient of variation (the ratio of standard deviation to mean) of the meteorological observation sequence is used to characterize the stability of meteorological changes. The larger the coefficient of variation, the more unstable the changes. The fluctuation amplitude and stability are normalized and mapped, and then structured and integrated according to the meteorological element type to generate a meteorological feature set.

[0066] S1.4 Extract hydrological parameters from the preprocessed hydrological process data, perform hydrological response analysis, and generate a hydrological feature set.

[0067] It should be noted that hydrological parameters such as flow rate, velocity, water level, and runoff are extracted from the preprocessed hydrological process data. The extracted hydrological parameters are then synchronized, aligned, and interpolated to form a hydrological parameter sequence. Trend decomposition is performed on the hydrological parameter sequence to extract the baseflow component and runoff component. The flow rate change, lag time, and peak duration of the hydrological parameter sequence are used as response parameters to characterize the response intensity of the hydrological process to external disturbances (such as rainfall or upstream replenishment). The response intensity, baseflow component, and runoff component are then normalized and integrated to generate a hydrological feature set.

[0068] It should also be noted that the baseflow component represents the long-term stable flow formed by groundwater recharge or the slow release of water from storage within the basin, reflecting the basic hydrological supply status of the basin; the runoff component represents the rapid response flow affected by rainfall, upstream confluence or surface runoff, reflecting the instantaneous contribution of short-term hydrological events to the river flow.

[0069] S1.5. Spatial location encoding is performed on the preprocessed geospatial data to generate a spatial feature set.

[0070] It should be noted that monitoring section coordinate data, river network node location data, topographic elevation data, and watershed boundary data are extracted from the preprocessed geospatial data. The monitoring section coordinate data and river network node location data are projected into a unified coordinate system and standardized to form a basic geospatial dataset. The node numbers of the river network node location data are used as indexes, and a spatial adjacency matrix is ​​constructed by combining the spatial distance and flow direction between adjacent nodes in the river network node location data. The topographic elevation data and watershed boundary data are spatially encoded to generate a spatial location index table. The basic geospatial dataset, spatial adjacency matrix, and spatial location index table are encoded and integrated to generate a spatial feature set.

[0071] S1.6 Analyze the aggregated fluctuations of the pretreated water quality physicochemical parameter data and output the environmental response feature set.

[0072] It should be noted that the pre-processed water quality physicochemical parameters are aligned and interpolated for missing data according to the time sequence of collection to form a water quality physicochemical parameter sequence. The water quality physicochemical parameter sequence is then time-aggregated to obtain the mean and variance of each water quality physicochemical parameter within the time window, which is used to reflect the overall change level of the water body's physicochemical properties. Wavelet decomposition and energy spectrum analysis are performed on the time-aggregated water quality physicochemical parameter sequence to identify the main fluctuation frequency bands and energy concentration intervals, which are used to characterize the sensitivity of the parameters to external disturbances. The time-aggregated results of each parameter, the main fluctuation frequency bands, and the energy concentration intervals are normalized and integrated to generate an environmental response feature set.

[0073] It should also be noted that the length of the time window is determined based on the monitoring data sampling period and the water body chemical reaction response period, such as 7–14 days.

[0074] S1.7 Perform baseline correction and migration analysis on the preprocessed isotope observation data to obtain the time series feature set.

[0075] It should be noted that hydrogen and oxygen isotope data, carbon and nitrogen isotope data, and sulfur isotope data are extracted from the preprocessed isotope observation data. The extracted isotope observation data are reconstructed and interpolated over time to form an isotope observation sequence. The moving average of the isotope observation sequence is used as the background trend value and integrated in chronological order to generate an isotope baseline sequence. The isotope observation sequence and the isotope baseline sequence are differentially analyzed to obtain an isotope shift sequence, which is used to reflect the dynamic changes in the isotope composition of the sample relative to the regional baseline. The isotope baseline sequence and the isotope shift sequence are normalized and integrated to generate a time-series feature set.

[0076] S1.8. The meteorological feature set, hydrological feature set, spatial feature set, environmental response feature set and temporal feature set are structurally integrated to generate a spatiotemporal environmental feature set.

[0077] It should be noted that the meteorological feature set, hydrological feature set, spatial feature set, environmental response feature set, and temporal feature set are aligned and indexed to achieve a one-to-one correspondence between features from different sources in the temporal and spatial dimensions, generating a multi-source feature set. The multi-source feature set is then subjected to Z-score standardization to unify the dimensions and numerical ranges of each feature to a comparable scale. A linear affine mapping is then performed on the standardized multi-source feature set through a feature mapping function, and the standardized multi-source feature set is encoded into a unified feature embedding space to generate a spatiotemporal environmental feature set, which is used to describe the comprehensive spatiotemporal response relationship between meteorology, hydrology, environment, and isotope changes.

[0078] S2. Input the spatiotemporal environmental feature set and the preprocessed measured isotope data into the isotope background model. The isotope background model performs time-series dependency modeling and spatial correlation inference to generate an isotope residual distribution map.

[0079] S2.1 Initialize the model parameters based on the spatiotemporal environment feature set, and use preprocessed measured isotope data for supervised training to construct an isotope background model;

[0080] It should be noted that the meteorological feature set, hydrological feature set, spatial feature set, environmental response feature set and temporal feature set in the spatiotemporal environmental feature set are uniformly indexed according to monitoring sections and time steps, and the features of each section under the same time step are spliced ​​to form a two-dimensional time step matrix. The two-dimensional time step matrix is ​​stacked in chronological order to generate a three-dimensional spatiotemporal feature tensor. The three-dimensional spatiotemporal feature tensor is normalized and decomposed into principal components to generate a standardized spatiotemporal feature matrix.

[0081] The standardized spatiotemporal feature matrix is ​​linearly combined with local adjacent nodes in the spatial dimension to generate a spatial correlation weight matrix; a sliding window linear transformation is performed in the temporal dimension to generate a temporal dependency mapping matrix; the spatial correlation weight matrix and the temporal dependency mapping matrix are weighted and fused in the feature dimension to generate a feature coupling representation matrix, which expresses the spatiotemporal coupling relationship between environmental factors and isotopic features; by parameter mapping of the feature coupling representation matrix, the weight coefficients of each feature dimension in the feature coupling representation matrix are decomposed and recombined to extract the isotopic background model parameter set representing spatial influence, temporal dependency and feature interaction; normalization and linear transformation are performed on the isotopic background model parameter set to construct the isotopic background model.

[0082] The isotope background model employs a hybrid spatial-temporal structure. The spatial dimension is implemented using a graph convolutional neural network (GNN), where the number of layers and kernel size are determined based on the number of surface water monitoring sections. For example, the GNN is set to have 2 layers and a kernel size of 3×3. The temporal dimension is implemented using a long short-term memory (LSTM) network, where the window length is determined based on the isotope observation period, such as 10. The feature fusion part generates a feature coupling representation matrix through a nonlinear mapping between a fully connected layer and a ReLU activation function. The training phase of the isotope background model uses the Adam optimization algorithm with a learning rate range of 0.0001 to 0.01. The upper limit of 0.01 is derived from the theoretical limit of convergence stability, while the lower limit of 0.0001 is derived from the empirical limit of convergence efficiency.

[0083] The preprocessed measured isotope data is used as supervised samples to input into the isotope background model. Forward prediction is performed on the standardized spatiotemporal feature matrix to generate predicted isotope values. The error between the predicted isotope values ​​and the measured isotope data is used as the optimization objective. The parameter set of the isotope background model (including spatial combination parameters, time transformation parameters, and feature fusion weights) is adjusted iteratively through gradient updates until the error of the isotope background model no longer decreases. The trained isotope background model is then output.

[0084] S2.2 Input the spatiotemporal environmental feature set into the isotopic background model, analyze the temporal correlation and spatial dependence of the spatiotemporal environmental feature set, and generate the expected isotopic value;

[0085] It should be noted that the spatiotemporal environmental feature set is input into the isotopic background model, and the spatiotemporal environmental feature set is expanded in the time dimension and reconstructed in the spatial dimension to form a spatiotemporal feature tensor. The spatiotemporal feature tensor is subjected to sliding window convolution in the time dimension to capture the temporal dependence features of meteorological and hydrological changes. In the spatial dimension, adjacent nodes in the spatial correlation weight matrix are weighted linearly to generate spatial diffusion features. The temporal dependence features and spatial diffusion features are combined and normalized to generate a comprehensive spatiotemporal response feature table. The comprehensive spatiotemporal response feature table is subjected to linear mapping and nonlinear activation to generate the expected isotope value.

[0086] S2.3. Compare and analyze the preprocessed measured isotope data with the expected isotope values, and output the isotope residual data.

[0087] It should be noted that the preprocessed measured isotope data and expected isotope values ​​are matched one-to-one with a unified time index to form an aligned dataset. Differential operations are performed on the aligned dataset according to the isotope channels to obtain the isotope residual sequence for each time step. The isotope residual sequence is subjected to sliding smoothing and scale unification processing to reduce measurement jitter and maintain comparability between channels. The processed isotope residual sequence is recombined into isotope residual data according to the time index.

[0088] S2.4 Perform spatial clustering and pattern recognition on the isotopic residual data to generate an isotopic residual distribution map.

[0089] It should be noted that isotopic residual data is paired with monitoring section coordinate data to generate a residual point set; the DBSCAN clustering algorithm is used, with the average spacing of the monitoring section and the spatial distribution density of residual points as neighborhood scale references, to perform connected component identification and merging of adjacent regions on the residual spatial point set, generating a candidate residual patch layer; boundary tracking and morphological correction are performed on the candidate residual patch layer to form a residual patch boundary set; residual statistical features such as area, perimeter, principal axis direction, and duration of each residual patch in the residual patch boundary set are extracted, and all residual statistical features are integrated to generate a residual pattern label table; the residual patch boundary set and the residual pattern label table are overlaid and visualized to output an isotopic residual distribution map.

[0090] S3. Perform spatial clustering analysis on the isotope residual distribution map to obtain a time evolution characteristic table, perform spatiotemporal coding on the time evolution characteristic table, and obtain the sequence of abnormal events.

[0091] S3.1 Perform residual intensity aggregation and connected domain division on the isotope residual distribution map to obtain the set of isotope anomaly regions.

[0092] It should be noted that the residual values ​​of each neighborhood in the isotope residual distribution map are aggregated and rank-fused by sliding window convolution to generate a residual intensity surface. The window size of the sliding window convolution is set according to the monitoring section spacing. Connectivity marking and merging of adjacent areas are performed on the residual intensity surface to form a candidate continuous area layer. Contour tracing and topology correction are performed on the candidate continuous area layer to output a set of isotope anomaly regions.

[0093] S3.2 Analyze the spatial correlation of the isotopic anomaly region set and output a table of temporal evolution characteristics.

[0094] It should be noted that morphological parameters such as the spatial center, side length, and area of ​​each anomalous region in the isotopic anomaly region set are extracted, and the contact boundary ratio and distance correlation between each anomalous region are obtained through the spatial adjacency matrix. The morphological parameters, contact boundary ratio, and distance correlation are integrated to generate a spatial relationship matrix. In the spatial relationship matrix, regions with continuous spatial correlation are identified, and regional evolution chains are established by combining time index information. The regional evolution chains are then processed by time sequence organization and relationship encoding to generate a time evolution feature table.

[0095] S3.3 Identify abnormal fluctuations in the time evolution feature table and perform time-series sorting to obtain the abnormal event window table.

[0096] It should be noted that the feature records in the time evolution feature table are sorted according to the time index to form a continuous time series; a sliding window smoothing process is performed on the continuous time series to remove local noise. The length of the current window is determined according to the record resolution of the time evolution feature table, such as 3 to 5 time steps. At the same time, the rate of change and trend direction information are extracted; fluctuation segments (the rate of change deviates significantly from the average rate of change and the trend direction is inconsistent) are detected on the smoothed continuous time series and the corresponding time intervals are marked, and an abnormal event window table is output.

[0097] S3.4 Perform spatial clustering and event persistence analysis on the abnormal event window table to generate an intensity feature table.

[0098] It should be noted that the abnormal event windows in the abnormal event window table are clustered according to the spatial location index and divided into multiple spatial cluster groups. The spatial concentration and overlap area ratio of abnormal events in the spatial cluster groups are used as spatial clustering features. The duration and frequency of occurrence of abnormal event windows are counted in each spatial cluster group to generate time persistence parameters. The spatial clustering features and time persistence parameters are structurally integrated to generate an intensity feature table.

[0099] S3.5 Analyze the spatiotemporal correlation of the intensity feature table and perform serialization processing to output the abnormal event sequence.

[0100] It should be noted that the process involves reading the time index, spatial location index, and intensity field (including spatial clustering, average residual amplitude, and duration of the abnormal event) from the intensity feature table; performing continuous comparisons on each record (the intensity field of a single abnormal event) in the intensity feature table according to the time index and marking adjacent record pairs; performing nearest neighbor matching within adjacent record pairs based on the spatial location index, and summarizing to form an event association table; sorting the event association table by time index and concatenating segments according to the connection relationship to output the abnormal event sequence.

[0101] S4. Analyze the temporal continuity and spatial variation of abnormal event sequences, output a pollution propagation feature table, perform spatiotemporal similarity matching and node association on the pollution propagation feature table, and generate a pollution propagation association matrix.

[0102] S4.1 Analyze the temporal continuity of the abnormal event sequence and generate an abnormal event feature table.

[0103] It should be noted that, based on the time interval and intensity change rate of adjacent abnormal events in the abnormal event sequence (obtained by the ratio of the intensity field to time), abnormal event segments with continuous growth or decay trends are identified; the duration, direction of change, and average intensity of the abnormal event segments are structurally integrated to generate an abnormal event feature table.

[0104] S4.2 Identify the spatial clustering of abnormal event sequences and perform spatial change trend analysis, outputting a spatial change feature table.

[0105] It should be noted that the spatial location index, spatial center coordinates, and intensity fields of each abnormal event in the abnormal event sequence are extracted; the spatial distance between adjacent abnormal events is obtained based on the spatial location index, and the abnormal events are aggregated according to the principle of minimizing spatial distance to form spatial clustered event groups; within each spatial clustered event group, spatial trajectory fitting is performed based on the spatial center coordinates to obtain the movement direction and position change of the center point, which are used as the direction vector and change amplitude; the average spatial distance, direction vector, and change amplitude of the spatial clustered event group are used as spatial clustering parameters, and the spatial clustering parameters of all spatial clustered event groups are summarized to generate a spatial change feature table.

[0106] S4.3. Index and match the abnormal event feature table with the spatial change feature table to obtain the spatiotemporal feature joint table. Perform spatiotemporal correlation analysis on the spatiotemporal feature joint table to generate the pollution transmission feature table.

[0107] It should be noted that, based on the time index and spatial location index, the abnormal event feature table and the spatial change feature table are matched using a dual-index system. Feature records with the same time period and spatial range are merged to generate a spatiotemporal feature joint table. From the spatiotemporal feature joint table, abnormal fields such as duration, intensity, direction vector, and change amplitude of abnormal events are extracted. The abnormal fields are normalized, and a first-order difference operation is performed on the normalized abnormal fields to obtain the time change rate and spatial migration rate. The abnormal fields, time change rate, and spatial migration rate are then structurally integrated to output the pollution propagation feature table.

[0108] S4.4 Calculate the spatiotemporal similarity of the pollution transmission characteristic table, perform event matching based on the spatiotemporal similarity, and output the spatiotemporal matching result table.

[0109] It should be noted that the spatiotemporal similarity of all abnormal events in the pollution propagation characteristic table is calculated, and the spatiotemporal similarity is sorted from high to low. The sorting results are then deduplicated and self-matched to ensure that each abnormal event is only associated with the most relevant other abnormal events. The temporal and spatial indices of adjacent abnormal event pairs are checked for continuity, and temporally continuous and spatially adjacent abnormal event pairs are combined into initial matching pairs. The time interval and spatial distance of the initial matching pairs are smoothed using linear interpolation to form event matching pairs with continuous propagation characteristics. All event matching pairs are then integrated to generate a spatiotemporal matching result table.

[0110] The expression for calculating the spatiotemporal similarity of all anomalous events in the pollution propagation characteristic table is as follows:

[0111] ;

[0112] in, Indicates an abnormal event and abnormal events The spatiotemporal similarity between them Indicates an abnormal event Time index, Indicates an abnormal event Time index, Indicates an abnormal event and abnormal events Spatial distance between them Represents the time characteristic scale. Indicates the scale of spatial characteristics. Indicates directional similarity. Indicates the time weighting factor. Indicates the spatial weighting coefficients. Represents the direction weighting coefficients, satisfying .

[0113] It should also be noted that the time weighting coefficient is determined by analyzing the variance of the time index of abnormal events in the pollution transmission characteristic table. The value range is 0.3 to 0.5, where the lower limit of 0.3 represents the minimum perceptible impact of time change on the similarity of pollution events, and the upper limit of 0.5 represents the upper limit weight when time continuity becomes the main influencing factor, so as to ensure that time factors and spatial factors are balanced in similarity calculation.

[0114] Based on the spatial coordinate field of abnormal events in the pollution transmission characteristic table, the spatial distance distribution between abnormal events is obtained. The inverse ratio of the standard deviation of spatial distance is used as the spatial weighting coefficient, with a value range of 0.3 to 0.5. The lower limit of 0.3 represents the case of loose spatial distribution and weak spatial influence, while the upper limit of 0.5 represents the dominant weight of spatial factors when abnormal events spread or gather continuously in space, thereby balancing the influence of geographical distance and time difference.

[0115] Based on the abnormal event propagation direction field in the pollution propagation characteristic table, the cosine of the angle between the direction vectors of all abnormal event pairs is obtained. The mean of the cosine of the direction vector angle is used as the direction weighting coefficient, with a value range of 0.1 to 0.3. The lower limit of 0.1 reflects the case where the directionality of the abnormal event is not obvious or the diffusion is strong, while the upper limit of 0.3 represents the maximum direction contribution when the direction is stable and the abnormal event propagates along a fixed path, ensuring that the direction factor maintains an auxiliary role in the overall spatiotemporal similarity calculation.

[0116] The temporal characteristic scale is determined by the difference between the time indices in the pollution propagation characteristic table, and is used to characterize the typical time span of the interval between abnormal events; the spatial characteristic scale is obtained by statistically analyzing the nearest neighbor distance distribution of the center coordinates of abnormal events in the spatial location index table data, and is used to reflect the typical scale of the spatial propagation of abnormal events.

[0117] Directional similarity is obtained by calculating the cosine similarity of the propagation direction vectors of each abnormal event in the spatial variation feature table, and is used to measure the consistency of the propagation direction of events. Directional similarity comes from the cosine value of the angle between the propagation direction vectors of all abnormal events in the spatial variation feature table, and is used to measure the consistency of the propagation direction of events.

[0118] S4.5 Perform nearest neighbor search and spatiotemporal feature fusion on the spatiotemporal matching result table to obtain the pollution transmission correlation matrix.

[0119] It should be noted that, based on the temporal order and spatial coordinates of each abnormal event in the spatiotemporal matching result table, a nearest neighbor search is performed to integrate the nearest neighbor events with the shortest spatial distance and highest similarity for each abnormal event, generating a neighbor event index table; the time interval and spatial distance of the abnormal events are extracted from the neighbor event index table, and the time interval and spatial distance are standardized to unify the units; using the abnormal event number as the row and column index, the time interval and spatial distance of all abnormal event pairs are extracted from the pollution propagation feature table, and the variances of all time intervals and all spatial distances are used as temporal fusion coefficients and spatial fusion coefficients, respectively. The standardized time intervals and spatial distances are weighted and fused according to the temporal fusion coefficients and spatial fusion coefficients to output the pollution propagation correlation matrix.

[0120] S5. Derive the link direction of the pollution propagation correlation matrix, generate a pollution propagation path set, perform reverse tracing and regional compilation on the pollution propagation path set, and output a surface water pollution source tracing report.

[0121] S5.1 Perform time-series weighting and node sorting on the pollution transmission correlation matrix to obtain the time-series link table.

[0122] The pollution propagation correlation matrix and the spatiotemporal matching result table are mapped by time index according to the abnormal event number to obtain relation entries with time indexes. Based on the chronological order, an exponential decay function is used to perform time-progressive weighting and forward filtering on the relation entries, retaining forward relations whose occurrence time is earlier than that pointing to the abnormal event. The retained forward relations are sorted by the abnormal event time index, and the start and end times and cumulative correlation values ​​of each forward event are summarized. The sorted forward relations are concatenated segment by segment in chronological order to generate a time-series link table containing the source event, target event, time sequence number, and intra-segment sequence position.

[0123] S5.2 Perform directional consistency filtering and spatial path reconstruction on the temporal link table to generate an initial propagation path set.

[0124] It should be noted that the spatial center coordinates of the source and target events of each link in the temporal link table are extracted to generate a link direction vector table; within consecutive links of the same temporal order, deviating links are filtered out according to the principle of minimum directional deviation, with the main direction vector as a reference, and a directional consistency link table is output; adjacent links in the directional consistency link table are spliced ​​together in chronological order to form candidate path polylines; topology deforking and redundant segment merging are performed on the candidate path polylines to retain the longest continuous segment, generating an initial propagation path set.

[0125] S5.3 Verify the continuity of the initial propagation path set and perform topology optimization to output the pollution propagation path set.

[0126] It should be noted that within each propagation path of the initial propagation path set, a continuity check is performed based on the quantile benchmark of the time index interval and spatial distance between adjacent links. Adjacent links whose time interval exceeds the time threshold are determined to be discontinuous segments. For discontinuous segments, nearby links are retrieved according to the spatiotemporal matching result table and plugged in to repair them, outputting a continuous repair path set. The continuous repair path set is then subjected to topology optimization, short branches and repeated loops are deleted, collinear adjacent segments are merged according to directional consistency, and vertices are constrained, aligned, and simplified according to the spatial position index table, outputting a pollution propagation path set.

[0127] It should also be noted that the time threshold is obtained by interval fitting of the mean and standard deviation of the global time interval distribution of adjacent links, with a value range of 14 hours to 18 hours. The 14-hour threshold comes from the statistical results of the time interval distribution of historical monitoring samples, while the 18-hour threshold comes from the delay and diffusion effects of physical processes such as surface runoff and underground recharge.

[0128] S5.4 Perform reverse path analysis and node backtracking on the pollution transmission path set, and output the source tracing node set.

[0129] It should be noted that, starting from the end node of each pollution transmission path in the pollution transmission path set, the upstream nodes in the path are traced back in reverse order of time index to form a reverse node chain. During the tracing process, based on the spatial orientation consistency of adjacent nodes in the pollution transmission association matrix, discontinuous or conflicting node pairs are filtered out, and the timestamps, spatial coordinates, and association weight information of the retained nodes are recorded. All reverse node chains are aggregated and merged according to spatial adjacency to output the source tracing node set.

[0130] S5.5. Compare the source node set with the pretreated water quality physicochemical parameter data using isotopic characteristics and perform correlation analysis with environmental parameters to generate a source area characteristic table.

[0131] It should be noted that, based on the node index of the source tracing node set, the isotopic composition parameters of each node are extracted from the preprocessed isotopic observation data; the isotopic composition parameters are paired with the preprocessed water quality physicochemical parameter data to identify the main driving features of isotopic changes in response to environmental factors; based on the topological relationship of nodes in the spatial adjacency matrix, nodes with high correlation are aggregated and all main driving features are statistically analyzed to output the source region feature table.

[0132] S5.6 Perform spatial clustering and regional boundary fitting on the source area feature table to obtain the pollution source area distribution map.

[0133] It should be noted that, based on the coordinate information and adjacency relationships of each node in the spatial location index table, the spatial adjacency between nodes is obtained; according to the spatial adjacency of adjacent nodes and the main driving features in the source area feature table, a spectral clustering algorithm is used to perform spatial clustering on the nodes with the maximization of inter-cluster distance and minimization of intra-cluster variance as the benchmark, generating a source area clustering result table; on the source area clustering result table, the spatial coordinates of the nodes in each cluster are fitted to the boundary and the contour is tracked to form continuous regional boundary data; the regional boundary data is overlaid and rendered with the geospatial basic dataset to generate a pollution source area distribution map.

[0134] S5.7 Overlay and compile the pollution source area distribution map with the pollution propagation path set to generate a surface water pollution source tracing report.

[0135] It should be noted that, based on the coordinate reference in the spatial location index table data, spatial coordinate registration and projection unification are performed on the pollution source area distribution map and pollution propagation path set to form a spatially aligned dataset. Overlay analysis of paths and source areas is performed on the spatially aligned dataset to identify the intersection nodes of path endpoints and source area boundaries, and the time index, isotopic characteristic parameters, and environmental response characteristics of the intersection nodes are extracted to generate a source tracing intersection table. The source tracing intersection table is then structurally integrated with the pollution propagation path set and the source area characteristic table to generate a surface water pollution source tracing report dataset. This dataset contains pollution propagation paths, key nodes, source area locations, and temporal evolution information, used to intuitively represent the comprehensive source tracing results of pollution sources, pollution paths, and spatiotemporal evolution processes.

[0136] In summary, this invention, by constructing an isotopic background model based on a spatiotemporal environmental feature set, performing time-series dependency modeling and spatial correlation inference on multi-source meteorological, hydrological, and environmental data, obtaining expected isotopic values, and generating isotopic residual distribution maps, effectively separates natural fluctuations from pollution disturbance signals, improving the accuracy and stability of anomalous isotope identification; and by constructing anomalous event sequences and analyzing pollution propagation characteristics, it achieves chain reconstruction and dynamic tracking of pollution events in the temporal and spatial dimensions, revealing pollution diffusion paths and source area characteristic distributions, and improving the accuracy and interpretability of surface water pollution source tracing.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for tracing the source of surface water pollution based on isotopes, characterized in that: include, Collect hydrological and meteorological data and measured isotope data and preprocess them to extract the spatiotemporal environmental feature set of the preprocessed hydrological and meteorological data; The spatiotemporal environmental feature set and preprocessed measured isotope data are input into the isotope background model. The isotope background model performs time-series dependency modeling and spatial correlation inference to generate an isotope residual distribution map. The specific steps are as follows. The model parameters are initialized based on the spatiotemporal environmental feature set, and supervised training is performed using preprocessed measured isotope data to construct an isotope background model. The spatiotemporal environmental feature set is input into the isotopic background model to analyze the temporal correlation and spatial dependence of the spatiotemporal environmental feature set, and generate the expected isotopic value. It should be noted that the spatiotemporal environmental feature set is input into the isotopic background model, and the spatiotemporal environmental feature set is expanded in the time dimension and reconstructed in the spatial dimension to form a spatiotemporal feature tensor. The spatiotemporal feature tensor is subjected to sliding window convolution in the time dimension to capture the temporal dependence features of meteorological and hydrological changes. In the spatial dimension, adjacent nodes in the spatial correlation weight matrix are weighted linearly combined to generate spatial diffusion features. The temporal dependence features and spatial diffusion features are spliced ​​and normalized to generate a comprehensive spatiotemporal response feature table. Linear mapping and nonlinear activation are performed on the comprehensive spatiotemporal response feature table to generate the desired isotope values; The preprocessed measured isotope data are compared and analyzed with the expected isotope values, and the isotope residual data are output. Spatial clustering and pattern recognition are performed on isotopic residual data to generate isotopic residual distribution maps; Spatial clustering analysis was performed on the isotope residual distribution map to obtain a time evolution feature table. Spatiotemporal coding was performed on the time evolution feature table to obtain the sequence of anomalous events. Analyze the temporal continuity and spatial variation of abnormal event sequences, output a pollution propagation feature table, perform spatiotemporal similarity matching and node association on the pollution propagation feature table, and generate a pollution propagation association matrix; The link directions of the pollution propagation correlation matrix are derived, a set of pollution propagation paths is generated, reverse tracing and regional compilation are performed on the pollution propagation path set, and a surface water pollution source tracing report is output.

2. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The hydrological and meteorological data include meteorological observation data, hydrological process data, geospatial data, water quality physicochemical parameter data, and isotope observation data; The preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

3. The isotope-based method for tracing the source of surface water pollution as described in claim 2, characterized in that: The specific steps for extracting the spatiotemporal environmental feature set from the preprocessed hydrological and meteorological data are as follows. By analyzing the fluctuation trends of preprocessed meteorological observation data, a meteorological feature set is generated; Hydrological parameters are extracted from the preprocessed hydrological process data, and hydrological response analysis is performed to generate a hydrological feature set. Spatial location encoding is performed on the preprocessed geospatial data to generate a spatial feature set; Analyze the aggregated fluctuations of pretreated water quality physicochemical parameter data and output an environmental response feature set; Baseline correction and migration analysis were performed on the preprocessed isotope observation data to obtain a time series feature set; The meteorological feature set, hydrological feature set, spatial feature set, environmental response feature set and temporal feature set are structurally integrated to generate a spatiotemporal environmental feature set.

4. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The specific steps for performing spatial clustering analysis on the isotope residual distribution map to obtain a time evolution characteristic table are as follows. Residual intensity aggregation and connected component partitioning are performed on the isotopic residual distribution map to obtain a set of isotopic anomaly regions. Analyze the spatial correlation of isotopic anomaly regions and output a table of temporal evolution characteristics.

5. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The specific steps for performing spatiotemporal encoding on the time evolution feature table to obtain the sequence of abnormal events are as follows. Identify abnormal fluctuations in the time evolution feature table and perform time-series sorting to obtain an abnormal event window table; Spatial clustering and event persistence analysis are performed on the abnormal event window table to generate an intensity feature table; Analyze the spatiotemporal correlation of the intensity feature table and perform serialization processing to output the abnormal event sequence.

6. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The analysis of the temporal continuity and spatial variation of abnormal event sequences, and the output of a pollution propagation characteristic table, are detailed in the following steps. Analyze the temporal continuity of abnormal event sequences and generate an abnormal event feature table; Identify the spatial clustering of abnormal event sequences and perform spatial variation trend analysis, outputting a spatial variation feature table; The abnormal event feature table and the spatial change feature table are indexed and matched to obtain a spatiotemporal feature joint table. Spatiotemporal correlation analysis is performed on the spatiotemporal feature joint table to generate a pollution transmission feature table.

7. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The specific steps for performing spatiotemporal similarity matching and node association on the pollution propagation feature table to generate a pollution propagation association matrix are as follows. Calculate the spatiotemporal similarity of the pollution transmission characteristic table, perform event matching based on the spatiotemporal similarity, and output the spatiotemporal matching result table; Nearest neighbor search and spatiotemporal feature fusion are performed on the spatiotemporal matching result table to obtain the pollution transmission correlation matrix.

8. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The specific steps for deriving the link directions of the pollution propagation correlation matrix and generating a set of pollution propagation paths are as follows. The pollution transmission correlation matrix is ​​processed by time-series weighting and node sorting to obtain a time-series link table. Perform directional consistency filtering and spatial path reconstruction on the temporal link table to generate an initial propagation path set; Verify the continuity of the initial propagation path set and perform topology optimization to output the pollution propagation path set.

9. The isotope-based method for tracing the source of surface water pollution as described in claim 1, characterized in that: The specific steps for performing reverse tracing and regional compilation of the pollution propagation path set to output a surface water pollution source tracing report are as follows. Perform reverse path analysis and node backtracking on the pollution transmission path set, and output the source tracing node set; The source area feature table is generated by comparing the isotopic characteristics of the source node set with the pretreated water quality physicochemical parameter data and analyzing the correlation with environmental parameters. Spatial clustering and regional boundary fitting are performed on the source area feature table to obtain a pollution source area distribution map; The pollution source area distribution map is overlaid with the pollution propagation path set to generate a surface water pollution source tracing report.

Images