Spatial-temporal spectrum strong correlation optimization method based on water ecological characteristic elements

Abstract

The invention provides a time-space spectrum strong correlation optimization method based on water ecological characteristic elements, and relates to the technical field of water ecological environment monitoring and simulation. Comprising five steps of synchronous acquisition and processing of multi-source water ecological data, spatial-temporal spectrum three-dimensional feature primitive structured construction, spatial-temporal spectrum strong association rule analysis based on a graph model, parameter optimization of a coupling physical mechanism and an association rule, and model closed-loop optimization under dual verification. According to the method, a standardized data set is generated by fusing multi-source data, space-time spectrum feature primitives are constructed, then space-time spectrum strong association rules among elements are mined by utilizing a graph attention network, then the rules are embedded into a mechanism model in a constraint form, parameter optimization is performed by adopting a hybrid optimization algorithm, dual driving of a physical mechanism and a data rule is realized, and the method is suitable for the spatial-time spectrum strong association rules. Finally, the optimization model is output through closed-loop verification iteration, and therefore the precision, rationality and generalization ability of water ecological simulation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of water ecological environment monitoring and simulation technology, and in particular to a spatiotemporal spectral strong correlation optimization method based on water ecological characteristic elements. Background Technology

[0002] With increasing emphasis on water environment protection, accurate spatiotemporal simulation and prediction of core water ecological characteristics (such as chlorophyll a concentration, suspended solids concentration, and water temperature) has become a key technical support for water resource management, eutrophication control, and ecological health assessment.

[0003] Currently, this mainstream technical approach mainly relies on two types of methods: the first is dynamic models based on physical, chemical, and biological process mechanisms, and the second is data-driven models based on statistical learning or machine learning. For the first type, it describes processes such as material transport, transformation, and biological growth by constructing differential equations, possessing clear physical meaning. However, it faces some problems in practical applications. For example, the models usually contain a large number of parameters that need calibration (such as algal growth rate, sedimentation coefficient, etc.), and the spatial heterogeneity and time-varying nature of these parameters make it difficult for traditional calibration methods to obtain the global optimal solution, and easily lead to "different parameters producing similar results," meaning that multiple sets of different parameters may produce similar simulation results, reducing the uniqueness and reliability of the model. On the other hand, the structure of mechanistic models is relatively fixed, making it difficult to fully absorb and express the complex and nonlinear spatiotemporal correlations reflected from massive amounts of observational data, especially the deep coupling relationships between different elements across time lags, spatial transmission, and spectral responses, resulting in limited model simulation capabilities for certain ecological phenomena (such as the sudden proliferation and dissipation of algal blooms).

[0004] For the second type, which learns patterns directly from historical observation data, they often perform well in terms of fitting accuracy. However, they typically exist as "black boxes," lacking explicit physical constraints. Their predictions sometimes violate basic ecological principles or the law of conservation of mass, resulting in poor interpretability. More importantly, the learning performance of pure data models heavily depends on the quantity and quality of training data. When faced with new scenarios not covered by the training data (such as extreme weather conditions or new pollution inputs), their extrapolation prediction performance may drop sharply, exhibiting insufficient stability and generalization ability.

[0005] Therefore, the construction of both mechanistic and data-driven models heavily depends on the quality of the input data. Currently, water ecological data sources are becoming increasingly diverse. Existing technologies, when processing these multi-source, multi-scale, and heterogeneous data, often employ simple assimilation or splicing methods, failing to achieve true spatiotemporal benchmark unification and deep information fusion. This results in the phenomenon of "data silos," preventing the formation of a complete, consistent, high-quality dataset for in-depth analysis and model-driven applications, thus limiting further improvements in model performance. Therefore, this invention proposes a spatiotemporal spectral strong correlation optimization method based on water ecological characteristic elements to address the problems existing in current technologies. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to propose a spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements. This invention can integrate the advantages of multi-source data, quantitatively mine the inherent spatiotemporal correlation laws of aquatic ecology, and organically combine these laws with physical mechanisms to solve the problems existing in the prior art.

[0007] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements, comprising the following steps:

[0008] Step 1: Synchronous Collection and Processing of Multi-Source Water Ecological Data

[0009] The system simultaneously collects time-series multispectral remote sensing images of the target water area, continuous monitoring data from fixed-site sensors, and periodic manual sampling data. Then, it performs radiometric calibration, atmospheric correction, and geometric fine correction on the time-series multispectral remote sensing images to generate standard surface reflectance and water surface temperature data. Subsequently, it unifies the data with the continuous monitoring data from fixed-site sensors and the periodic manual sampling data to the same geographic coordinate system and time reference, and then fuses them to generate a standardized dataset of chlorophyll a concentration, suspended matter concentration, and water temperature with unified spatiotemporal resolution.

[0010] Step 2: Structural Construction of Three-Dimensional Feature Primitives in Spatiotemporal Spectrum

[0011] Based on a standardized dataset, regular grids are defined as basic hydrological spatial units and fixed durations are defined as basic time units to construct a spatiotemporal spectrum three-dimensional data cube. Then, spatial neighborhood features, time series features, and spectral features are extracted from each spatial unit in the spatiotemporal spectrum three-dimensional data cube for each time unit, and this set of features is defined as a spatiotemporal spectrum feature primitive.

[0012] Step 3: Analysis of Spatiotemporal Spectral Strong Correlation Rules Based on Graph Model

[0013] Based on a standardized dataset, time series of various aquatic ecological elements are constructed. Then, the causal relationship and spatial clustering pattern between the time series of different aquatic ecological elements are analyzed. Using each spatiotemporal spectral feature primitive as a node, a spatiotemporal spectral association graph is constructed based on spatial adjacency, time series autoregression relationship and spectral feature similarity. Graph attention network is applied to the spatiotemporal spectral association graph to parse the quantitative association rules between aquatic ecological elements across spatiotemporal and spectral dimensions, thus forming a set of strong spatiotemporal spectral association rules.

[0014] Step 4: Parameter optimization of coupling physical mechanisms and association rules

[0015] The set of strongly correlated spatiotemporal rules is embedded in the mechanism model that couples the hydrodynamic module and the ecological dynamics module in the form of constraints. An objective function is constructed with the goal of minimizing the error between simulated and observed values ​​and the degree of violation of the correlation rules as the penalty term. A hybrid optimization framework combining the Lagrange multiplier method and the particle swarm optimization algorithm is used to iteratively optimize the key ecological dynamic parameters in the mechanism model.

[0016] Step 5: Model Closed-Loop Optimization under Dual Validation

[0017] The optimized model is applied to the independent validation period and the region for forward simulation. Quantitative validation is performed using independent observation data from three dimensions: numerical accuracy, spatial pattern similarity, and temporal dynamic consistency. Based on the validation results, the feature dimensions of the spatiotemporal spectral feature primitives are corrected, and steps four and five are repeated until the model performance reaches the preset standard, at which point the final optimized model is output.

[0018] A further improvement lies in the following: In step one, the specific method for fusing and generating the standardized dataset is as follows:

[0019] S1: A ground-based measured dataset consisting of discrete spatial point data is constructed by combining continuous monitoring data from fixed-site sensors with periodic manual sampling data.

[0020] S2: Then, from the standard surface reflectance and water surface temperature data, extract the pixel spectral information that corresponds to the ground-measured dataset in space and time;

[0021] S3: Based on the ground-based measured dataset and its corresponding pixel spectral information, establish a remote sensing inversion model for chlorophyll a concentration, suspended matter concentration and water temperature, and apply the model to generate inversion data for each element covering the continuous spatial area of ​​the target water body.

[0022] S4: Using the Kriging interpolation method, the ground-based measured dataset composed of discrete spatial point data is fused with the inversion data of the continuous coverage area to generate a standardized dataset.

[0023] Further improvements are made in the following steps: In step two, the spatial neighborhood features are the mean and standard deviation of the elements calculated using a 3x3 window, the time series features are the linear trend and seasonal amplitude obtained by calculating the data from the previous four weeks, and the spectral features are the ratio of the normalized vegetation index to the reflectance of a specific band.

[0024] A further improvement is made in step three, where the causal relationship is analyzed using the Granger causality test and the spatial autocorrelation analysis method is used to analyze the spatial clustering pattern.

[0025] A further improvement is that, in step three, constructing a spatiotemporal spectrum correlation graph based on spatial adjacency specifically refers to establishing connections between corresponding nodes based on the spatial unit adjacency relationships determined by the Thiessen polygon method.

[0026] A further improvement is that, in step three, constructing a spatiotemporal spectral correlation graph based on spectral feature similarity specifically means establishing a connection between corresponding nodes when the cosine similarity between the spectral feature vectors of two nodes is greater than a set threshold.

[0027] A further improvement is made in step four, where the ecological dynamics module is an algae growth model.

[0028] A further improvement is made in step five, which includes enhancing the spectral feature dimension.

[0029] The beneficial effects of this invention are as follows: By synchronously processing and spatially fusing remote sensing imagery, fixed-site data, and manually sampled data, this invention generates a standardized dataset with unified spatiotemporal benchmarks and resolutions, providing a high-quality and consistent data foundation for subsequent analysis. Furthermore, by constructing a spatiotemporal spectral three-dimensional data cube and extracting well-defined multidimensional feature primitives, it achieves an integrated representation of the state of the aquatic ecosystem in spatial, temporal, and spectral dimensions, constructing a structured carrier for in-depth exploration of its intrinsic correlations. Then, based on this, using the spatiotemporal spectral strong correlation rule set parsed by graph models and attention mechanisms, the implicit, cross-dimensional nonlinear interactions in the data are transformed into quantifiable and computable ones. The knowledge constraints were then addressed by embedding this rule set as a penalty term into the objective function of the mechanistic model. A hybrid optimization algorithm was then used to optimize the parameters, achieving dual driving force and mutual correction between physical mechanisms and statistical data. This ensured that while the model fits the observed values, its internal dynamics must also conform to the systematic correlations revealed by the data, thus significantly improving the physical rationality and overall accuracy of the simulation. Finally, through independent spatiotemporal verification and feature and model iteration based on performance feedback, a self-improving closed-loop optimization process was formed, ensuring that the final output model not only has high accuracy but also good generalization ability and stability, thereby providing a reliable technical tool for the precise management of the aquatic ecological environment. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the workflow of the present invention. Detailed Implementation

[0031] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0032] according to Figure 1 As shown, this embodiment proposes a spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements. In this embodiment, taking a target water area in 2023 as an example, the method includes the following steps:

[0033] Step 1: Synchronous Collection and Processing of Multi-Source Water Ecological Data

[0034] Simultaneously, 30 Landsat-8 satellite images and 40 Sentinel-2 satellite images covering the target water area in 2023 were acquired. Hourly sensor data from three fixed automatic water quality monitoring stations within the area were also acquired, including data on water temperature, chlorophyll a fluorescence value, and turbidity. Additionally, laboratory analysis data from manual sampling conducted on the 15th of each month at 10 pre-defined sampling points in the area were acquired, analyzing chlorophyll a concentration, suspended solids concentration, and water temperature.

[0035] Then, radiometric calibration, atmospheric correction based on the MODTRAN model, and geometric fine correction for registration with the Taihu Lake Basin Digital Elevation Model were performed on all 70 satellite images. The processing generated standard surface reflectance products (including blue, green, red, and near-infrared bands) with a spatial resolution of 30 meters, as well as water surface temperature products derived from the radiative transfer equation.

[0036] Then, the plane coordinates of all monitoring station data and sampling point data were unified to the CGCS2000 coordinate system, the elevation datum was unified to the 1985 National Elevation Datum, and the time was unified to Beijing time. The hourly data of the monitoring stations were aggregated into the average value of the day (±12 hours) of the image transit, and the manually sampled data were associated with the nearest available image date on the 15th of each month.

[0037] Based on the above data fusion, a standardized dataset is generated, and the specific method is as follows:

[0038] S1: Combine sensor data from fixed monitoring stations with laboratory data from manual sampling points to form a ground-based measured dataset consisting of discrete spatial point data. This dataset contains measured values ​​of chlorophyll a concentration, suspended matter concentration, and water temperature at each sampling time at 13 spatial points.

[0039] S2: For each record in the ground-based measured dataset, based on its spatial coordinates and recording date, extract the blue light band reflectance, green light band reflectance, red light band reflectance, near-infrared band reflectance and water surface temperature value of the pixel at that coordinate from the corresponding standard surface reflectance and water surface temperature data products to form the pixel spectral information corresponding to that record;

[0040] S3: Using the measured chlorophyll a concentration from the ground-based dataset as the dependent variable, and the corresponding red band reflectance, near-infrared band reflectance, and their ratio as the main independent variables, a random forest regression algorithm was used to train the chlorophyll a concentration remote sensing inversion model. Similarly, using the measured suspended matter concentration as the dependent variable and the red band reflectance as the independent variable, a linear inversion model was established. Then, using the measured water temperature as the dependent variable and the water surface temperature retrieved from the satellite as the independent variable, a linear correction model was established. Finally, these three models were applied to perform pixel-by-pixel calculations on all 70 processed images to generate spatially continuous chlorophyll a concentration inversion maps, suspended matter concentration inversion maps, and water temperature inversion maps covering the entire Meiliang Bay water area.

[0041] S4: A standard Kriging interpolation method is used for fusion. This involves using measured data from 13 discrete spatial points as the primary observations, and then using the spatial variation structure (characterized by a variogram) provided by the previously generated spatial continuous inversion maps of the three elements as the basis for spatial correlation in the interpolation. For each target output grid (set to 100m × 100m), the Kriging algorithm uses the measured point data and the spatial trends provided by the inversion maps to estimate the chlorophyll a concentration, suspended solids concentration, and water temperature values ​​for that grid in each week. This ultimately generates a standardized dataset with a spatiotemporal resolution of weekly 100m grids. This dataset contains weekly spatial grid data of the three element concentrations for the target water area.

[0042] Step 2: Structural Construction of Three-Dimensional Feature Primitives in Spatiotemporal Spectrum

[0043] Based on the standardized dataset generated in step one, the target water area was divided into a regular grid of 100m × 100m, resulting in 1250 spatial units. Then, using "week" as the basic time unit, the data from week 1 to week 52 of 2023 were arranged in chronological order. The chlorophyll a concentration, suspended matter concentration, and water temperature data for each spatial unit for 52 weeks, as well as the NDVI index and green to blue band ratio data calculated from the reflectance data for the corresponding weeks, were stacked together to form a spatiotemporal spectral three-dimensional data cube with a structure of 1250 (spatial) × 52 (time) × 5 (element / spectral index).

[0044] For each spatial unit in the data cube, the following three types of features are extracted week by week along its time dimension to form a spatiotemporal spectral feature primitive:

[0045] Spatial neighborhood characteristics: Taking the current spatial unit as the center, take the chlorophyll a concentration values ​​of the same cycle within the surrounding 3×3 grid (i.e., a range of 300m×300m). Calculate the arithmetic mean of these 9 values ​​as the local mean, and calculate its standard deviation as the local spatial variability.

[0046] Time series characteristics: Extract the chlorophyll a concentration value series of the current spatial unit for four consecutive weeks before this week (i.e., the first week before the fourth week before), perform univariate linear regression on the series, and the slope obtained is the linear trend. Calculate the absolute difference between the mean of the series and the mean of the series at this point throughout the year as a proxy indicator of seasonal amplitude.

[0047] Spectral characteristics: The Normalized Difference Vegetation Index (NDVI) is calculated using the near-infrared and red reflectance of the current week, and the ratio (G / B) is calculated using the green and blue reflectance of the current week.

[0048] Each spatiotemporal spectral feature primitive is a set containing the above five specific feature values ​​(spatial mean, spatial standard deviation, temporal trend, seasonal amplitude, NDVI, and G / B ratio), corresponding to the state of a specific spatial unit in a specific cycle.

[0049] Step 3: Analysis of Spatiotemporal Spectral Strong Correlation Rules Based on Graph Model

[0050] From the standardized dataset, weekly time series of chlorophyll a concentration and water temperature for the entire year of 2023 were extracted from all spatial units within the target water area. Then, the Granger causality test was used, with a lag of one week, to examine whether the water temperature time series was a Granger cause of the chlorophyll a concentration time series. In multiple spatial units in the northern region of the target water area, the test results showed p-values ​​less than 0.05, confirming that water temperature changes have a statistically significant leading indicator effect on chlorophyll a changes. Simultaneously, the global Moran index was used to analyze the spatial autocorrelation of chlorophyll a concentration at the end of each week; the index values ​​ranged from 0.2 to 0.5, indicating significant spatial clustering.

[0051] Therefore, the 1250 (spatial) × 52 (temporal) = 65000 spatiotemporal spectral feature primitives generated in step two are each treated as a graph node, and then connections (edges) between nodes are established according to the following three rules:

[0052] Spatial adjacency: The 1250 spatial units are divided based on the Thiessen polygon method. If the Thiessen polygons of two spatial units have a common boundary, they are determined to be spatially adjacent units, and all time nodes (52 pairs in total) belonging to these two units are connected pairwise.

[0053] Autoregressive relationship in time series: For the same spatial unit, a directed connection is established between the node in week t and the nodes in the previous week 1 and 2 weeks to represent the temporal dependency.

[0054] Spectral feature similarity: Calculate the cosine similarity between the spectral feature vectors (i.e., the two-dimensional vector composed of the NDVI value and G / B ratio of the node) of any two nodes (from different cycles and different spatial units). Set the similarity threshold to 0.85. If the calculated cosine similarity is greater than 0.85, then an undirected connection is established between the two nodes.

[0055] Therefore, a spatiotemporal spectral correlation graph is constructed using each spatiotemporal spectral feature primitive as a node, based on spatial adjacency, time-series autoregressive relationships, and spectral feature similarity. This graph is then input into a two-layer graph attention network. The input layer of this network is the feature vector of each node, and the network learns the importance weights of each edge (i.e., connections between nodes) in the graph through a multi-head attention mechanism. After training, the network can assign different attention coefficients to different types of connections (such as spatial adjacency edges, time-lag edges, and hyperspectral similarity edges). By decoding these coefficients and the node states, quantitative correlation rules can be extracted, such as "when a node's chlorophyll a concentration is low in the previous two weeks and the current water temperature of one of its spatially adjacent nodes is high, the probability of the node's chlorophyll a concentration increasing next week increases significantly." Finally, a set of strong correlation rules covering various spatiotemporal spectral scenarios and their confidence levels are output.

[0056] Step 4: Parameter optimization of coupling physical mechanisms and association rules

[0057] A mechanistic model suitable for shallow lakes is constructed, coupling a hydrodynamic module based on shallow water equations and an ecological module based on algal growth dynamics. The core of the ecological module is the algal growth model, whose expression includes key parameters such as maximum algal growth rate, basal respiration rate, sedimentation rate, and half-saturated light intensity. Then, from the rule set obtained in step three, rules with a confidence level higher than 0.9 are selected and transformed into mathematical constraints. For example, a rule about "a positive correlation between high water temperature in spatially adjacent areas and subsequent increases in local chlorophyll a" is transformed into a constraint that the correlation coefficient between "water temperature in spatially adjacent areas and local chlorophyll a concentration" simulated by the model should not be lower than a certain threshold R.

[0058] Then, the objective function J is constructed, and its expression is:

[0059]

[0060] In the formula, The root mean square error between the spatiotemporal field of chlorophyll a simulated by the model and the standardized dataset. The weighting coefficient for the rule penalty item. (Rule violation) is the sum of squares of the differences between the actual simulated relation strength and the expected relation strength of all embedded rules.

[0061] Furthermore, the inner loop uses the Lagrange multiplier method to introduce the rule constraints into the objective function in the form of multipliers, performing local gradient search on continuous parameters such as the maximum algal growth rate and respiration rate to quickly approximate the local optimum that satisfies the current constraints. The outer loop uses the particle swarm optimization algorithm, initializing 50 particles, each representing a complete set of model parameters (including...). PSO updates the particle's velocity and position based on its own historical best position and the historical best position of the entire population, searching in the global parameter space. Every 10 iterations, the current optimal parameters found by PSO are passed to the Lagrange multiplier method for fine-grained local optimization, and the results are then fed back to PSO to update the population.

[0062] After 200 iterations, the algorithm converges, outputting a set of optimal parameter combinations that minimize the objective function J, including the maximum algal growth rate of 1.2 / day, the settling rate of 0.15 m / day, and the weights of the rule penalty terms. .

[0063] Step 5: Model Closed-Loop Optimization under Dual Validation

[0064] The optimized mechanism model was applied to an independent verification period (in this example, 2023, so the verification period was set from July 1 to September 30, 2023, during which the data did not participate in any preliminary modeling) and an independent sub-region of the target water area (the grid in this region did not participate in the spatial similarity calculation in the rule mining).

[0065] Using independent monitoring data from monitoring stations during this period, as well as two additional manual sampling data points for the area, quantitative verification was conducted from three dimensions: numerical accuracy, spatial pattern similarity, and temporal dynamic consistency.

[0066] Numerical accuracy: The correlation coefficient between the simulated chlorophyll a concentration time series and the data from the MLW-03 monitoring station was 0.78, and the root mean square error was 3.2 μg / L;

[0067] Spatial pattern similarity: Comparing the simulation results on August 15 with the Sentinel-2 image inversion results on the same day, the spatial distribution pattern similarity index of the two reached 0.65;

[0068] Temporal dynamic consistency: The model successfully simulated the spatial redistribution of algae driven by wind in late August, which is consistent with the observed direction of movement of the high chlorophyll a value area.

[0069] Validation revealed that the model's simulation of chlorophyll a response during a rapid warming event in early July exhibited a lag of approximately 3 days. Analysis indicated that the current feature primitives insufficiently captured the spectral response features for short-term thermal events. Therefore, we returned to step two, adding a "recent water temperature change rate" feature to the spatiotemporal spectral feature primitives. This feature is calculated from the difference between the current week's water temperature and the previous week's water temperature. Then, using the enhanced feature primitives, we re-executed the graphical model training in step three (using only data from before July), resulting in a set of supplementary association rules incorporating the effects of short-term water temperature changes.

[0070] The supplementary rules were then embedded into the model, and the optimization process in step four was repeated (with only parameter fine-tuning). The updated model was used to simulate the validation period again, and the chlorophyll a response lag was improved, with the correlation coefficient with the MLW-03 data increasing to 0.82. After evaluation, the model performance stabilized and met the predetermined standards. Finally, this solidified model was output for aquatic ecosystem simulation and prediction in the target water area.

[0071] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A spatiotemporal spectral strong correlation optimization method based on water ecological characteristic elements, characterized by: Includes the following steps: Step 1: Synchronous Collection and Processing of Multi-Source Water Ecological Data The system simultaneously collects time-series multispectral remote sensing images of the target water area, continuous monitoring data from fixed-site sensors, and periodic manual sampling data. Then, it performs radiometric calibration, atmospheric correction, and geometric fine correction on the time-series multispectral remote sensing images to generate standard surface reflectance and water surface temperature data. Subsequently, it unifies the data with the continuous monitoring data from fixed-site sensors and the periodic manual sampling data to the same geographic coordinate system and time reference, and then fuses them to generate a standardized dataset of chlorophyll a concentration, suspended matter concentration, and water temperature with unified spatiotemporal resolution. Step 2: Structural Construction of Three-Dimensional Feature Primitives in Spatiotemporal Spectrum Based on a standardized dataset, regular grids are defined as basic hydrological spatial units and fixed durations are defined as basic time units to construct a spatiotemporal spectrum three-dimensional data cube. Then, spatial neighborhood features, time series features, and spectral features are extracted from each spatial unit in the spatiotemporal spectrum three-dimensional data cube for each time unit, and this set of features is defined as a spatiotemporal spectrum feature primitive. Step 3: Analysis of Spatiotemporal Spectral Strong Correlation Rules Based on Graph Model Based on a standardized dataset, time series of various aquatic ecological elements are constructed. Then, the causal relationship and spatial clustering pattern between the time series of different aquatic ecological elements are analyzed. Using each spatiotemporal spectral feature primitive as a node, a spatiotemporal spectral association graph is constructed based on spatial adjacency, time series autoregression relationship and spectral feature similarity. Graph attention network is applied to the spatiotemporal spectral association graph to parse the quantitative association rules between aquatic ecological elements across spatiotemporal and spectral dimensions, thus forming a set of strong spatiotemporal spectral association rules. Step 4: Parameter optimization of coupling physical mechanisms and association rules The set of strongly correlated spatiotemporal rules is embedded in the mechanism model that couples the hydrodynamic module and the ecological dynamics module in the form of constraints. An objective function is constructed with the goal of minimizing the error between simulated and observed values ​​and the degree of violation of the correlation rules as the penalty term. A hybrid optimization framework combining the Lagrange multiplier method and the particle swarm optimization algorithm is used to iteratively optimize the key ecological dynamic parameters in the mechanism model. Step 5: Model Closed-Loop Optimization under Dual Validation The optimized model is applied to the independent validation period and the region for forward simulation. Quantitative validation is performed using independent observation data from three dimensions: numerical accuracy, spatial pattern similarity, and temporal dynamic consistency. Based on the validation results, the feature dimensions of the spatiotemporal spectral feature primitives are corrected, and steps four and five are repeated until the model performance reaches the preset standard, at which point the final optimized model is output.

2. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step one, the specific method for fusing and generating a standardized dataset is as follows: S1: A ground-based measured dataset consisting of discrete spatial point data is constructed by combining continuous monitoring data from fixed-site sensors with periodic manual sampling data. S2: Then, from the standard surface reflectance and water surface temperature data, extract the pixel spectral information that corresponds to the ground-measured dataset in space and time; S3: Based on the ground-based measured dataset and its corresponding pixel spectral information, establish a remote sensing inversion model for chlorophyll a concentration, suspended matter concentration and water temperature, and apply the model to generate inversion data for each element covering the continuous spatial area of ​​the target water body. S4: Using the Kriging interpolation method, the ground-based measured dataset composed of discrete spatial point data is fused with the inversion data of the continuous coverage area to generate a standardized dataset.

3. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step two, the spatial neighborhood features are the mean and standard deviation of the elements calculated using a 3x3 window, the time series features are the linear trend and seasonal amplitude obtained by calculating the data from the previous four weeks, and the spectral features are the ratio of the normalized vegetation index to the reflectance of a specific band.

4. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step three, the causal orientation relationship is analyzed using the Granger causality test, and the spatial autocorrelation analysis method is used to analyze the spatial clustering pattern.

5. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step three, constructing a spatiotemporal spectrum association graph based on spatial adjacency specifically refers to establishing connections between corresponding nodes based on the spatial unit adjacency relationships determined by the Thiessen polygon method.

6. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step three, constructing a spatiotemporal spectral correlation graph based on spectral feature similarity specifically means establishing a connection between corresponding nodes when the cosine similarity between the spectral feature vectors of two nodes is greater than a set threshold.

7. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step four, the ecological dynamics module is an algae growth model.

8. The spatiotemporal spectral strong correlation optimization method based on aquatic ecological characteristic elements according to claim 1, characterized in that: In step five, the correction includes enhancing the spectral feature dimension.

Images