Dynamic monitoring method for cross-seasonal seawater salinity abnormity

Abstract

The invention relates to the technical field of environmental monitoring, in particular to a cross-seasonal seawater salinity anomaly dynamic monitoring method, which comprises the following steps: S1, collecting multi-source driving factors, constructing a standardized multi-source driving factor set, generating a regional coupling structure map through a dynamic structure identification mechanism, and marking a key boundary; s2, taking the region coupling structure atlas as input, combining measured salinity time sequence data, constructing a salinity response model, switching a weight function, and outputting a dynamic salinity response estimation result with region coupling adaptation capability; and S3, based on a dynamic salinity response estimation result, constructing an abnormal evolution path map, and forming a cross-seasonal salinity anomaly monitoring result oriented to a causal link. By constructing a cross-seasonal regional coupling structure map, heterogeneity information of multi-source driving factors is effectively fused, a weighted graph structure is constructed by adopting a sliding seasonal window and correlation constraints, dynamic evolution characteristics of a hydrological-climate coupling relationship are accurately identified, and the problem of insufficient identification capability is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a method for dynamic monitoring of seawater salinity anomalies across seasons. Background Technology

[0002] Against the backdrop of marine environmental changes, the spatiotemporal dynamics of salinity in nearshore and estuarine areas are being driven by a variety of hydrological and climatic factors. During seasonal changes, factors such as wind direction, estuarine runoff, tidal amplitude, sea surface temperature, and local circulation are coupled with each other, significantly affecting the transport pathways and diffusion mechanisms of seawater salinity. The sudden occurrence and evolution of salinity anomalies not only reflect changes in the dynamic stability of regional ecosystems but are also important precursors to environmental risks such as seawater intrusion and red tide outbreaks. Therefore, constructing a cross-seasonal salinity anomaly monitoring method with spatiotemporal coupling sensing capabilities has significant scientific research and management implications.

[0003] Existing methods for monitoring salinity anomalies mostly focus on static data fitting, single-factor driven modeling, or fixed threshold judgment, making it difficult to dynamically characterize the salinity change response brought about by seasonal coupling mode switching. In addition, when faced with complex hydrological structure changes and the combined effects of multiple regional factors, traditional methods lack the ability to model the evolution mechanism of the coupling structure, resulting in the inability to effectively capture the spatial propagation path of salinity anomalies and their potential chain-like causal relationships, making it difficult to support the identification of the starting node of abnormal events and the quantitative analysis of response delay. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a method for dynamic monitoring of seawater salinity anomalies across seasons. This method involves constructing a regional structural map reflecting the coupling characteristics of hydrology and climate, integrating salinity response modeling and structural adaptation mechanisms, to accurately estimate the dynamic response of salinity under coupling mode switching, and further combining spatial propagation path identification and anomaly chain transmission mechanisms to output monitoring results including anomaly level, initiation factor, and response delay.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a method for dynamic monitoring of seawater salinity anomalies across seasons, comprising the following steps: S1: Collect multi-source driving factors of the target sea area in different seasons, construct a standardized multi-source driving factor set, generate a regional coupling structure map reflecting the hydrological-climate coupling state through a dynamic structure recognition mechanism, and mark the key boundaries of coupling mode switching. S2, taking the regional coupling structure map as input, combined with measured salinity time series data, a salinity response model is constructed using a structure adaptation mapping algorithm, the weight function is automatically switched under different coupling states, and a dynamic salinity response estimation result with regional coupling adaptation capability is output. S3. Based on the dynamic salinity response estimation results and combined with the trend of coupling structure changes, identify the spatial propagation path and triggering chain nodes of salinity changes, construct an anomaly evolution path map, and output the anomaly level, initiation factor and response delay information accordingly, forming a cross-seasonal salinity anomaly monitoring result oriented towards causal links.

[0006] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein S1 includes: S11 collects multi-source driving factors of the target sea area in different seasons, including wind direction, estuary runoff, tidal amplitude, sea surface temperature and local circulation structure, and performs unified formatting and temporal-spatial alignment on the collected multi-source driving factors to construct a standardized multi-source driving factor set. S12, divide the driving factors into sliding seasonal windows, and standardize the variables in each window to eliminate dimensional differences. S13, within each seasonal window, calculate the Pearson correlation coefficient between each driving factor and generate a correlation matrix reflecting the coupling relationship between the factors; S14. Based on the correlation matrix, construct a weighted graph structure with driving factors as nodes, retain edges with coupling strength higher than the coupling strength screening threshold, and form a regional coupling structure map representing the coupling mechanism of the current season. S15 quantifies the structural changes in the regional coupling structure map of consecutive seasons, identifies the time boundaries of abrupt changes in coupling relationships, and marks them as key boundaries for coupling mode switching.

[0007] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein step S14 includes: S141 uses the wind direction, estuary runoff, tidal amplitude, sea surface temperature, and local circulation structure within the current seasonal window as nodes in the graph, forming a set of driving factor nodes. ,in, Let i be the i-th driving factor, i=1,2,…M, and M be the total number of driving factors; S142, based on the calculated correlation matrix, driving factor pairs with correlation strength higher than the coupling strength screening threshold are extracted as edges in the graph. Driving factor connections with coupling relationships are retained, and interfering edges are removed, as shown below: ; ; in, The set of edges to be retained. As driving factor and There are coupling edges between them. For coupling strength, Let be the Pearson correlation coefficient between factors i and j in the k-th season. The threshold for selecting coupling strength; S143 combines nodes and filtered edges into an undirected weighted graph. The weight of each edge is determined by the absolute value of the Pearson correlation coefficient of the driving factors, forming a regional coupling structure map that reflects the current seasonal coupling mechanism. ,in, For the edge The weights represent the coupling strength. The graph is an undirected graph, i.e., if... ,but ,and .

[0008] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein S15 includes: S151 compares the regional coupling structure maps of adjacent seasons, statistically analyzes the average change in the edge weights between driving factors, quantifies the intensity of changes in the map structure, and reflects the overall fluctuation of the coupling relationship, expressed as: ; in, The intensity of graph structure change between the k-th and k+1-th seasonal windows. Let be the set of all edges that appear in both graphs. Let (a, b) be the weight of edge (a, b) in the graph of the k-th season; if it does not appear, set it to 0. This represents the number of edges involved in the change calculation. , For the driving factor node index, a≠b; S152, compare the intensity of graph structure change with the structural mutation threshold. If it exceeds the structural mutation threshold, the current time point is determined to be the coupling mode switching boundary, indicating that the coupling mechanism has undergone a mutation. S153, summarize all identified switching time points to generate a set of critical boundaries for coupled mode switching. ,in, This refers to the time point at which abrupt changes in the coupling structure occur between the k-th seasonal window and the (k+1)-th seasonal window. This is the threshold for structural mutation.

[0009] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein S2 includes: S21, embed the regional coupling structure map corresponding to each season into the model and extract the structural state vector representing the coupling relationship characteristics; S22. Based on the actual collected salinity time series data, calculate the average salinity and rate of change in the season, and construct a salinity response model. S23. Based on the structural state vector extracted from the regional coupling structure map, dynamically match the salinity response model weights that are compatible with the current coupling state, and apply them to the salinity response model to estimate the seasonal salinity response value, and output the prediction result with coupling structure adaptability.

[0010] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein step S21 includes: S211. For the regional coupling structure map of each season, extract the node and edge weight information, and construct the corresponding weighted adjacency matrix to represent the coupling strength between driving factors. , , ;in, For nodes and The coupling strength between them It is a set of edges representing coupling relationships; S212 assigns an initial feature vector to each driving factor node in the regional coupling structure graph, forming an initial node feature matrix. ,in, For nodes The initial feature vector, p=1,2,…R, where R is the number of nodes; S213: The weighted adjacency matrix and the initial node feature matrix are input into the graph neural network. The embedding representation of each node is extracted using graph convolution, and then aggregated into a structure state vector using full graph pooling. As a holistic expression of the current seasonal coupling structure, , ,in, The graph embedding results for each node, To add a self-loop to the adjacency matrix, The corresponding degree matrix is ​​defined as follows: W represents the learnable weights of the GCN. This is the ReLU activation function.

[0011] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein step S22 includes: S221, within the current seasonal window, calculates the salinity state vector based on the actual collected salinity time-series data, including the mean salinity and the rate of change, which respectively reflect the overall salinity level and trend of the sea area. S222, taking the salinity state vector as input, defines an adjustable-weight linear response model as the salinity response model, expressed as: ,in, The salinity response is predicted for season v. For weight parameters, This is a bias term used to estimate the salinity response value under the current season.

[0012] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein step S23 includes: S231, based on the structural state vector extracted for each season, construct a weight mapping function to dynamically generate weights and bias terms for the salinity response model that match the current structural state, expressed as: , ,in, This represents the dynamic weight vector of the salinity response model. This is the bias term for the salinity response model. , These are the first-level mapping matrix and the bias, respectively. , These are the corresponding bias mapping parameters. Let w be the structural state vector of the w-th season window; S232, the weights and biases dynamically generated from the structural state vector are applied to the constructed salinity response model to perform a linear transformation on the salinity state vector of the current season, thereby obtaining the predicted salinity response value. This serves as a prediction result with the ability to adapt to coupled structures.

[0013] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein S3 includes: S31. By calculating the deviation between the current seasonal salinity response value and the historical average, and combining the fluctuation amplitude for standardized judgment, the starting time window of the cross-seasonal abnormal response is identified, and then it is determined as the abnormal response trigger point. S32, based on the seasonal regional coupling structure map, analyzes the transmission mechanism of abnormal events among multiple driving factors. By analyzing the node coupling strength in the regional coupling structure map and the succession relationship of abnormal events in adjacent seasons, it constructs the abnormal propagation path and captures the potential chain-like trigger structure. S33, based on the abnormal propagation path, identify the initial trigger node in the entire abnormal link and calculate the propagation delay of each downstream response node relative to the starting node. ; S34, combining the two dimensions of abnormal response magnitude and propagation delay, calculates the abnormality level score of each node, and generates a monitoring result set including abnormality initiation factor, abnormal propagation path, propagation delay and abnormality level score.

[0014] The above-mentioned dynamic monitoring method for cross-seasonal seawater salinity anomalies, wherein step S32 includes: S321, based on the intra-seasonal node coupling strength and cross-seasonal anomaly correlation, calculate the anomaly transmission potential score between each driving factor; S322: Construct anomaly propagation paths using edges whose anomaly propagation potential scores are higher than the anomaly propagation score threshold, and identify chain paths with anomaly propagation relationships between multiple nodes. S323, analyze the cumulative influence of the starting node in the abnormal propagation path, and identify potential sources of abnormal propagation and key triggering nodes.

[0015] The beneficial effects of the present invention's dynamic monitoring method for cross-seasonal seawater salinity anomalies are as follows: By constructing a cross-seasonal regional coupling structure map, it effectively integrates heterogeneous information from multiple driving factors such as wind direction, estuary runoff, tidal amplitude, sea surface temperature, and local circulation. Furthermore, by constructing a weighted graph structure using a sliding seasonal window and correlation constraints, it can accurately identify the dynamic evolution characteristics of hydrological-climate coupling relationships, thereby extracting key boundaries for coupling mode switching and achieving high-sensitivity detection of abrupt changes in marine environmental structure. This effectively compensates for the insufficient ability of existing technologies to identify inter-seasonal structural transitions.

[0016] By introducing graph embedding and structure adaptation mapping mechanisms, a structure state vector is generated using coupled structure graphs and linked with salinity state data to construct a dynamic salinity response model with structure-driven characteristics. The model can automatically adjust parameter weights according to the coupling state in different seasons, realizing structure-adaptive modeling and dynamic prediction of salinity response. This solves the problem of decreased response accuracy of existing models under complex and changing coupling relationships, and significantly improves the model's generalization ability and robustness.

[0017] By guiding the causal relationship modeling of abnormal salinity response and structural propagation path, and further constructing an anomaly propagation map based on the anomaly transmission potential, the system identifies key inducing nodes, propagation paths and response delays, and outputs anomaly level scores oriented towards chain mechanisms. This enables causal link analysis and propagation tracing of cross-seasonal anomaly events, effectively supporting the accurate monitoring and scientific early warning of marine salinity anomalies, and significantly enhancing the system's ability to understand and respond to sudden anomalies. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the monitoring method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the construction of the regional coupling structure map according to an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be described below in conjunction with specific embodiments and accompanying drawings.

[0020] Example 1 like Figures 1-2 As shown, a method for dynamic monitoring of seawater salinity anomalies across seasons includes the following steps: S1 collects multi-source driving factors of the target sea area in different seasons, including wind direction, estuary runoff, tidal amplitude, sea surface temperature and local circulation structure. Through a dynamic structure recognition mechanism, it generates a regional coupling structure map reflecting the hydrological-climate coupling state and marks the key boundaries of coupling mode switching.

[0021] include: S11 collects multi-source driving factors of the target sea area in different seasons, including wind direction, estuary runoff, tidal amplitude, sea surface temperature and local circulation structure, and performs unified formatting and temporal-spatial alignment on the collected multi-source driving factors to construct a standardized multi-source driving factor set, which specifically includes the following aspects.

[0022] (1) Collect multi-source driving factors from different data sources, including: Wind direction: from satellite wind vector product.

[0023] Estuarine runoff: from regional hydrological stations or land surface simulation systems.

[0024] Tidal amplitude: derived from coastal tide gauge stations or FES / TPXO tidal models.

[0025] Sea surface temperature: from MODIS or AVHRR remote sensing data.

[0026] Local circulation structure: derived from high-resolution ocean circulation models (such as ROMS, HYCOM) or observation buoy arrays.

[0027] (2) Align all driving factors along a unified time axis and fill in any missing data. Let the unified time series be... For each driving factor The original observation time series Perform time resampling and missing data interpolation.

[0028] Unified as: ,in, Let m be the original observation value of the m-th driving factor at time t. To unify the standardized value of the driving factor at time t, Indicates to The interpolation function at time t uses linear interpolation.

[0029] (3) Let the target space grid be The resolution is , the original driving factor Projected onto a standard grid using bilinear interpolation.

[0030] Represented as: ,in, These are the driving factor values ​​after reprojection. (x, y) represents the coordinate position in the original driving factor, and (x, y) represents the position on the target standard grid.

[0031] (4) Organize the unified multi-source driving factor set into a four-dimensional tensor, and output the standardized multi-source driving factor set, represented as: m = 1, 2, ..., M, where, For a standardized set of multi-source driving factors, (x, y) represents spatial grid coordinates, t represents the time index, and m represents the factor number. For example... Wind direction, Estuary runoff, Tidal amplitude, Sea surface temperature, Local circulation velocity magnitude.

[0032] S12 divides the driving factors into sliding seasonal windows and standardizes the variables within each window to eliminate dimensional differences.

[0033] Specifically, it includes: (1) Let the unified time series be Divide it into lengths The sliding window is divided into multiple sub-windows. Each window is defined as: ,in, For the k-th sliding seasonal window, Let k be the starting time of the k-th window. The time span for each window, and K is the total number of sliding windows.

[0034] (2) For each driving factor In each window Within, Z-score normalization is performed, and it is represented as: ,in, These are the standardized driving factor values. , The m-th factor in the window Mean and standard deviation of the positions (x, y) within the range.

[0035] S13, within each seasonal window, calculate the Pearson correlation coefficients between the driving factors, and generate a correlation matrix reflecting the coupling relationship between the factors, as shown below: , ,in, As driving factor and The correlation strength, Let be the coupling factor correlation matrix of the k-th season window at the spatial location (x, y).

[0036] S14. Based on the correlation matrix, construct a weighted graph structure with driving factors as nodes, retain only the edges with coupling strength higher than the coupling strength screening threshold, and form a regional coupling structure map representing the coupling mechanism of the current season.

[0037] include: S141 uses the wind direction, estuary runoff, tidal amplitude, sea surface temperature, and local circulation structure within the current seasonal window as nodes in the graph, forming a set of driving factor nodes. ,in, Let i be the i-th driving factor, i = 1, 2, ..., M, where M is the total number of driving factors.

[0038] S142, based on the calculated correlation matrix, driving factor pairs with correlation strength higher than the coupling strength screening threshold are extracted as edges in the graph. Only driving factor connections with coupling relationships are retained, and interfering edges are removed, as shown below: ; ; in, The set of edges to be retained. As driving factor and There are coupling edges between them. For coupling strength, Let be the Pearson correlation coefficient between factors i and j in the k-th season. The threshold for selecting coupling strength.

[0039] S143 combines nodes and filtered edges into an undirected weighted graph. The weight of each edge is determined by the absolute value of the Pearson correlation coefficient of the driving factors, forming a regional coupling structure map that reflects the current seasonal coupling mechanism. ,in, For the edge The weights represent the coupling strength. The graph is an undirected graph, i.e., if... ,but ,and .

[0040] S15 quantifies the structural changes in the regional coupling structure map of consecutive seasons, identifies the time boundaries of abrupt changes in coupling relationships, and marks them as key boundaries for coupling mode switching.

[0041] include: S151 compares the regional coupling structure maps of adjacent seasons, statistically analyzes the average change in the edge weights between driving factors, quantifies the intensity of changes in the map structure, and reflects the overall fluctuation of the coupling relationship, expressed as: ; in, The intensity of graph structure change between the k-th and k+1-th seasonal windows. Let be the set of all edges that appear in both graphs. Let (a, b) be the weight of edge (a, b) in the graph of the k-th season; if it does not appear, set it to 0. This represents the number of edges involved in the change calculation. , For the driving factor node index, a≠b.

[0042] S152, compare the intensity of the graph structure change with the structural mutation threshold. If it exceeds the structural mutation threshold, the current time point is determined to be the coupling mode switching boundary, indicating that the coupling mechanism has undergone a mutation.

[0043] The structural mutation threshold is expressed as: ,in, This is the threshold for structural mutation. , These represent the mean and standard deviation of all structural change intensity indices, respectively. This represents the mutation sensitivity coefficient.

[0044] S153, summarize all identified switching time points to generate a set of critical boundaries for coupled mode switching. , is represented as: ,in, This represents the time point at which abrupt changes in the coupling structure occur between the k-th seasonal window and the (k+1)-th seasonal window.

[0045] S2 takes the regional coupling structure map as input, combines it with measured salinity time series data, and uses the structure-adaptive mapping algorithm to construct a salinity response model. It automatically switches the weight function under different coupling states and outputs dynamic salinity response estimation results with regional coupling adaptation capability.

[0046] include: S21, embed the regional coupling structure map corresponding to each season into the model and extract the structural state vector representing the coupling relationship.

[0047] include: S211: For the regional coupling structure graph of each season, extract the node and edge weight information and construct the corresponding weighted adjacency matrix. , used to represent the coupling strength between driving factors, is expressed as: , ;in, For nodes and The coupling strength between edges (i.e., edge weights). This is the set of edges representing coupling relationships.

[0048] S212 assigns an initial feature vector to each driving factor node in the regional coupling structure graph, forming an initial node feature matrix. , is represented as: ,in, For nodes The initial feature vectors are p=1,2,…R, where R is the number of nodes.

[0049] S213: The weighted adjacency matrix and the initial node feature matrix are input into the graph neural network. The embedding representation of each node is extracted using graph convolution, and then aggregated into a structure state vector using full graph pooling. As a holistic expression of the current seasonal coupling structure, it is represented as: ,in, The graph embedding results for each node, To add a self-loop to the adjacency matrix, The corresponding degree matrix is ​​defined as follows: W represents the learnable weights of the GCN. This is the ReLU activation function.

[0050] .

[0051] S22. Based on the actual collected salinity time-series data, calculate the average salinity and rate of change during the season, and construct a salinity response model.

[0052] include: S221, within the current seasonal window, based on the actual collected salinity time-series data, calculate the salinity state vector, including the mean salinity and the rate of change, reflecting the overall salinity level and trend of the sea area, respectively, as follows: , , ,in, This represents the average salinity. Let be the salinity observation value at the i-th sampling time, and be the total number of samples within the current seasonal window. The rate of change of salinity, This is the salinity state vector.

[0053] S222, taking the salinity state vector as input, defines an adjustable-weighted linear response model as the salinity response model to estimate the salinity response value under the current season, expressed as: ,in, The salinity response is predicted for season v. For weight parameters, This is a bias term used to estimate the salinity response value under the current season.

[0054] S23. Based on the structural state vector extracted from the regional coupling structure map, dynamically match the salinity response model weights that are compatible with the current coupling state, and apply them to the salinity response model to estimate the seasonal salinity response value, and output the prediction result with coupling structure adaptability.

[0055] include: S231, based on the structural state vector extracted for each season, construct a weight mapping function to dynamically generate weights and bias terms for the salinity response model that match the current structural state, expressed as: , ,in, This represents the dynamic weight vector of the salinity response model. This is the bias term for the salinity response model. , These are the first-level mapping matrix and the bias, respectively. , These are the corresponding bias mapping parameters. Let w be the structural state vector of the w-th season window.

[0056] S232, the weights and biases dynamically generated from the structural state vector are applied to the constructed salinity response model to perform a linear transformation on the salinity state vector of the current season, thereby obtaining the predicted salinity response value. This serves as a prediction result with the ability to adapt to coupled structures.

[0057] S3, based on the dynamic salinity response estimation results and combined with the trend of coupling structure changes, identifies the spatial propagation path and triggering chain nodes of salinity changes, constructs an anomaly evolution path map, and outputs anomaly level, initiation factor and response delay information accordingly, forming cross-seasonal salinity anomaly monitoring results oriented towards causal links.

[0058] include: S31, by calculating the deviation between the current seasonal salinity response value and the historical average, and combining this with the fluctuation amplitude for standardization, the starting time window of the cross-seasonal abnormal response is identified, and this is determined as the abnormal response trigger point, denoted as... , is represented as: ,in, For each seasonal window q, the salinity response value, This is the historical average. For historical standard deviation, If q is an abnormal sensitivity factor and the condition is met, then q is determined to be an abnormal response trigger point.

[0059] S32, based on the seasonal regional coupling structure map, analyzes the transmission mechanism of anomalous events among multiple driving factors. By analyzing the node coupling strength in the regional coupling structure map and the succession relationship of anomalous events in adjacent seasons, it constructs anomalous propagation paths and captures potential chain-like trigger structures.

[0060] include: S321, based on the intra-seasonal node coupling strength and cross-seasonal anomaly correlation, calculate the anomaly propagation potential score between each driving factor, expressed as: ,in, For season s from the driving factor node Towards The score of abnormal transmission potential, For nodes in the regional coupling structure map of season s and The coupling strength (i.e., edge weight). For nodes Does an abnormal event occur in season s? (1 if abnormal, 0 otherwise) For nodes Does the anomaly occur for the first time in the next season s+1 (1 if yes, 0 otherwise)?

[0061] S322, constructing anomaly propagation paths using edges whose anomaly propagation potential scores are higher than the anomaly propagation score threshold, identifies chain paths with anomaly propagation relationships between multiple nodes, represented as: ,in, This is a set of anomaly propagation paths, where path represents the anomaly propagation path. This is a directed graph of anomaly propagation. This is the threshold for abnormal transmission score.

[0062] The threshold for abnormal transmission score is expressed as: ; in, , , respectively, are the mean and standard deviation of the abnormal transmission potential scores for all edges.

[0063] S323 analyzes the cumulative influence of the starting node in the abnormal propagation path, identifies potential sources of abnormal propagation and key triggering nodes, represented as: ,in, The overall influence of node n as an anomaly initiation factor. The score represents the anomalous propagation potential from node n to node p in season s. Let n be a directed edge pointing from node n to node p.

[0064] S33, based on the abnormal propagation path, identify the initial trigger node in the entire abnormal link and calculate the propagation delay of each downstream response node relative to the starting node. , is represented as: ,in, For nodes The abnormal seasonal timestamp, For nodes The time point at which salinity anomalies occur.

[0065] S34, combining the two dimensions of abnormal response magnitude and propagation delay, calculates the abnormality level score for each node, generating a monitoring result set including the abnormality initiation factor, abnormal propagation path, propagation delay, and abnormality level score, represented as: ,in, Assess the level of abnormality. For nodes The salinity response value, The corresponding mean, , These are the weighting factors for response intensity and time decay, respectively.

[0066] This embodiment covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of this technical solution. To provide the public with a thorough understanding of the invention, specific details are described in detail in the embodiments; however, those skilled in the art can fully understand the invention even without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the invention, well-known methods, processes, flows, components, and circuits are not described in detail.

[0067] The above embodiments are merely illustrative of the structural concept and features of the present invention, intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring of seawater salinity anomalies across seasons, characterized in that, Includes the following steps: S1: Collect multi-source driving factors of the target sea area in different seasons, construct a standardized multi-source driving factor set, generate a regional coupling structure map reflecting the hydrological-climate coupling state through a dynamic structure recognition mechanism, and mark the key boundaries of coupling mode switching. S2, taking the regional coupling structure map as input, combined with measured salinity time series data, a salinity response model is constructed using a structure adaptation mapping algorithm, the weight function is automatically switched under different coupling states, and a dynamic salinity response estimation result with regional coupling adaptation capability is output. S3. Based on the dynamic salinity response estimation results and combined with the trend of coupling structure changes, identify the spatial propagation path and triggering chain nodes of salinity changes, construct an anomaly evolution path map, and output the anomaly level, initiation factor and response delay information accordingly, forming a cross-seasonal salinity anomaly monitoring result oriented towards causal links.

2. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 1, characterized in that, S1 includes: S11 collects multi-source driving factors of the target sea area in different seasons, including wind direction, estuary runoff, tidal amplitude, sea surface temperature and local circulation structure, and performs unified formatting and temporal-spatial alignment on the collected multi-source driving factors to construct a standardized multi-source driving factor set. S12 divides the driving factors into sliding seasonal windows and standardizes the variables in each window to eliminate dimensional differences. S13, within each seasonal window, calculate the Pearson correlation coefficient between each driving factor and generate a correlation matrix reflecting the coupling relationship between the factors; S14. Based on the correlation matrix, construct a weighted graph structure with driving factors as nodes, retain edges with coupling strength higher than the coupling strength screening threshold, and form a regional coupling structure map representing the coupling mechanism of the current season. S15 quantifies the structural changes in the regional coupling structure map of consecutive seasons, identifies the time boundaries of abrupt changes in coupling relationships, and marks them as key boundaries for coupling mode switching.

3. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 2, characterized in that, S14 includes: S141 uses the wind direction, estuary runoff, tidal amplitude, sea surface temperature, and local circulation structure within the current seasonal window as nodes in the graph, forming a set of driving factor nodes. ,in, Let i be the i-th driving factor, i=1,2,…M, and M be the total number of driving factors; S142, based on the calculated correlation matrix, driving factor pairs with correlation strength higher than the coupling strength screening threshold are extracted as edges in the graph. Driving factor connections with coupling relationships are retained, and interfering edges are removed, as shown below: ； ； in, The set of edges to be retained. As driving factor and There are coupling edges between them. For coupling strength, Let be the Pearson correlation coefficient between factors i and j in the k-th season. The threshold for selecting coupling strength; S143 combines nodes and filtered edges into an undirected weighted graph. The weight of each edge is determined by the absolute value of the Pearson correlation coefficient of the driving factors, forming a regional coupling structure map that reflects the current seasonal coupling mechanism. ,in, For the edge The weights represent the coupling strength. The graph is an undirected graph, i.e., if... ,but ,and .

4. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 3, characterized in that, S15 includes: S151 compares the regional coupling structure maps of adjacent seasons, statistically analyzes the average change in the edge weights between driving factors, quantifies the intensity of changes in the map structure, and reflects the overall fluctuation of the coupling relationship, expressed as: ； in, The intensity of graph structure change between the k-th and k+1-th seasonal windows. Let be the set of all edges that appear in both graphs. Let (a, b) be the weight of edge (a, b) in the graph of the k-th season; if it does not appear, set it to 0. This represents the number of edges involved in the change calculation. , For the driving factor node index, a≠b; S152, compare the intensity of graph structure change with the structural mutation threshold. If it exceeds the structural mutation threshold, the current time point is determined to be the coupling mode switching boundary, indicating that the coupling mechanism has undergone a mutation. S153, summarize all identified switching time points to generate a set of critical boundaries for coupled mode switching. ,in, This refers to the time point at which abrupt changes in the coupling structure occur between the k-th seasonal window and the (k+1)-th seasonal window. This is the threshold for structural mutation.

5. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 1, characterized in that, S2 includes: S21, embed the regional coupling structure map corresponding to each season into the model and extract the structural state vector representing the coupling relationship characteristics; S22. Based on the actual collected salinity time series data, calculate the average salinity and rate of change during the season, and construct a salinity response model. S23. Based on the structural state vector extracted from the regional coupling structure map, dynamically match the salinity response model weights that are compatible with the current coupling state, and apply them to the salinity response model to estimate the seasonal salinity response value, and output the prediction result with coupling structure adaptability.

6. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 5, characterized in that, S21 includes: S211. For the regional coupling structure map of each season, extract the node and edge weight information, and construct the corresponding weighted adjacency matrix to represent the coupling strength between driving factors. , ， ;in, For nodes and The coupling strength between them It is a set of edges representing coupling relationships; S212 assigns an initial feature vector to each driving factor node in the regional coupling structure graph, forming an initial node feature matrix. ,in, For nodes The initial feature vector, p=1,2,…R, where R is the number of nodes; S213: The weighted adjacency matrix and the initial node feature matrix are input into the graph neural network. The embedding representation of each node is extracted using graph convolution, and then aggregated into a structure state vector using full graph pooling. As a holistic expression of the current seasonal coupling structure, , ,in, The graph embedding results for each node, To add a self-loop to the adjacency matrix, The corresponding degree matrix is ​​defined as follows: W represents the learnable weights of the GCN. This is the ReLU activation function.

7. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 6, characterized in that, S22 includes: S221, within the current seasonal window, calculates the salinity state vector based on the actual collected salinity time-series data, including the mean salinity and the rate of change, which respectively reflect the overall salinity level and trend of the sea area. S222, taking the salinity state vector as input, defines an adjustable-weight linear response model as the salinity response model, expressed as: ,in, The salinity response is predicted for season v. For weight parameters, This is a bias term used to estimate the salinity response value under the current season.

8. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 7, characterized in that, S23 includes: S231, based on the structural state vector extracted for each season, construct a weight mapping function to dynamically generate weights and bias terms for the salinity response model that match the current structural state, expressed as: ， ,in, This represents the dynamic weight vector of the salinity response model. This is the bias term for the salinity response model. , These are the first-level mapping matrix and the bias, respectively. , These are the corresponding bias mapping parameters. Let w be the structural state vector of the w-th season window; S232, the weights and biases dynamically generated from the structural state vector are applied to the constructed salinity response model to perform a linear transformation on the salinity state vector of the current season, thereby obtaining the predicted salinity response value. This serves as a prediction result with the ability to adapt to coupled structures.

9. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 8, characterized in that, S3 includes: S31. By calculating the deviation between the current seasonal salinity response value and the historical average, and combining the fluctuation amplitude for standardized judgment, the starting time window of the cross-seasonal abnormal response is identified, and then it is determined as the abnormal response trigger point. S32, based on the seasonal regional coupling structure map, analyzes the transmission mechanism of abnormal events among multiple driving factors. By analyzing the node coupling strength in the regional coupling structure map and the succession relationship of abnormal events in adjacent seasons, it constructs the abnormal propagation path and captures the potential chain-like trigger structure. S33, based on the abnormal propagation path, identify the initial trigger node in the entire abnormal link and calculate the propagation delay of each downstream response node relative to the starting node. ; S34, combining the two dimensions of abnormal response magnitude and propagation delay, calculates the abnormality level score of each node, and generates a monitoring result set including abnormality initiation factor, abnormal propagation path, propagation delay and abnormality level score.

10. The method for dynamic monitoring of cross-seasonal seawater salinity anomalies according to claim 9, characterized in that, S32 includes: S321, based on the intra-seasonal node coupling strength and cross-seasonal anomaly correlation, calculate the anomaly transmission potential score between each driving factor; S322: Construct anomaly propagation paths using edges whose anomaly propagation potential scores are higher than the anomaly propagation score threshold, and identify chain paths with anomaly propagation relationships between multiple nodes. S323, analyze the cumulative influence of the starting node in the abnormal propagation path, and identify potential sources of abnormal propagation and key triggering nodes.

Images