A three-dimensional evaluation method for ecological restoration effect

Abstract

The present application provides a kind of ecological restoration effect stereoscopic evaluation method, belong to ecological restoration evaluation technical field, the present application is obtained by being laid out in target repair area multi-point monitoring network comprehensive basic ecological parameters, constructs dynamic threshold value ecological early warning algorithm system to realize abnormal mode intelligent identification and automatic early warning trigger, establishes multi-parameter coupling analysis mechanism and according to the dynamic adjustment monitoring processing frequency of vegetation cover value, designs eigenvalue sensitivity analysis mechanism and according to the repair control strategy of the change rate of adjacent matrix eigenvalue, constructs ecological succession trend prediction model based on time series graph convolution network, extracts key succession features using multi-head attention mechanism to predict future evolution trend, combined with 360 degree panoramic technology to establish stereoscopic display system before and after repair to generate comprehensive evaluation report, solve the technical problem that ecological restoration effect evaluation is not accurate enough.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of ecological restoration evaluation, and in particular relates to a stereoscopic evaluation method for ecological restoration effect. BACKGROUND

[0002] The evaluation of the ecological restoration effect of marine salt marshes is an important technical means in marine ecological protection and restoration engineering. Traditional ecological restoration effect evaluation mainly relies on fixed-point sampling monitoring, remote sensing image interpretation and statistical analysis methods for single-dimensional evaluation. By setting fixed monitoring sample points in the restoration area, basic ecological parameters such as vegetation coverage, biomass and species diversity are obtained, and simple mathematical statistical methods and comparative analysis techniques are used to quantitatively evaluate the ecological conditions before and after restoration. This method is widely used in the fields of marine beach ecological restoration, mangrove restoration engineering and salt marsh wetland reconstruction projects. Traditional techniques have technical defects such as insufficient spatial representativeness due to sparse monitoring points, inability to reflect dynamic changes through single time node evaluation, lack of analysis of internal relationships in ecological systems, and incomplete evaluation index system, which cannot accurately capture the complex spatio-temporal variation characteristics and interaction relationships between ecological elements in the ecological restoration process. The evaluation results often deviate from the actual restoration effect. In the current practice of marine ecological restoration evaluation, due to the lack of spatial continuity analysis capability, time series dynamic monitoring mechanism and multi-dimensional parameter coupling evaluation system in traditional methods, it is difficult to establish an accurate quantitative evaluation model of ecological restoration effect, resulting in low evaluation result accuracy, weak reliability and insufficient guidance. That is, there is a technical problem of inaccurate evaluation of ecological restoration effect in the prior art. SUMMARY

[0003] Therefore, the present application provides a stereoscopic evaluation method for ecological restoration effect, which can solve the technical problem of inaccurate evaluation of ecological restoration effect in the prior art.

[0004] The application is implemented in the following manner: the application provides a three-dimensional evaluation method for ecological restoration effect, a multi-point monitoring network is arranged in a target restoration area, remote sensing images and field investigation are combined to obtain basic ecological parameters of salt marsh vegetation such as species composition, restoration area, retained area, plant height, coverage, density, survival rate and biomass, and three-dimensional spatial coordinates and timestamp information of each monitoring point are recorded; based on the obtained discrete monitoring data, a Kriging interpolation and machine learning fusion algorithm is used to predict the spatial continuous distribution of the study area, the influence range of each discrete point is calculated by analyzing the spatial position correlation characteristics, the support vector regression model is used to correct the interpolation result, and a global ecological index distribution map is generated; an ocean ecological network connectivity analysis model is established, the ecological patches are abstracted as nodes, the ecological corridors are abstracted as edges, an ecological network graph is constructed, topological indexes such as network connectivity, intermediate centrality and clustering coefficient are calculated, key ecological nodes are identified through node importance sorting and network robustness analysis; a dynamic threshold ecological early warning algorithm system is constructed, a dynamic control limit of the ecological index is established combined with the control chart theory, an abnormal pattern is identified by using the support vector machine, and the sliding window technology is used to update the threshold parameter, and the early warning mechanism is automatically triggered when the monitoring value exceeds the control limit; a multi-parameter coupling analysis mechanism is established, the monitoring processing frequency is adjusted according to the salt marsh vegetation coverage value; a characteristic value sensitivity analysis mechanism is designed, the sensitivity of the eigenvalues of the ecological network adjacency matrix to the change of the environmental stress parameter is calculated, and the ecological restoration control strategy is adjusted according to the sensitivity analysis result; an ecological succession trend prediction model is constructed, a time series graph convolution network architecture is used to process spatio-temporal sequence data, key ecological succession features are extracted by using a multi-head attention mechanism, the attention weight distribution is dynamically adjusted according to three key parameters of network connectivity, vegetation coverage and biodiversity index, the future ecological restoration evolution trend is predicted and output.

[0005] In the step of arranging the multi-point monitoring network, a spatially uniformly distributed monitoring point grid is established in the target restoration area, macro ecological condition information is obtained by combining remote sensing image technology, detailed vegetation ecological parameters are collected by field investigation, the spatial representativeness and time continuity of the monitoring data are ensured, and basic data support is provided for subsequent spatial interpolation analysis and ecological network construction.

[0006] The basic ecological parameters are specific quantitative ecological indexes of salt marsh vegetation species composition, restoration area, retained area, plant height, coverage, density, survival rate and biomass obtained by monitoring, and three-dimensional spatial coordinates and timestamp information of each monitoring point are recorded to form a complete spatio-temporal ecological data set, which provides a data basis for quantitative evaluation of ecological restoration effect.

[0007] Specifically, the ecological network adjacency matrix refers to a two-dimensional matrix that describes the connection relationships between nodes in an ecological network. The matrix element values ​​represent the connection strength or connection status between nodes. The changes in the matrix eigenvalues ​​reflect the stability of the ecological network structure and the degree of impact of environmental stress on the ecosystem.

[0008] Before constructing the ecological network map, the process includes preprocessing steps such as accurately identifying the boundaries of ecological patches and assessing the connectivity of ecological corridors. Spatial analysis techniques are used to determine the geometric shape and size of ecological patches and identify the width and connectivity of ecological corridors, providing accurate spatial information for the abstract modeling of nodes and edges of the ecological network.

[0009] Specifically, the weight allocation parameters of the multi-head attention mechanism are dynamically calculated based on three parameters: network connectivity, vegetation cover, and biodiversity index obtained from current monitoring. When network connectivity is in different intervals, the attention weight is tilted towards spatial features; when vegetation cover is in different intervals, the attention weight is tilted towards temporal features; and when the biodiversity index is in different intervals, the attention weight is evenly distributed between spatiotemporal features.

[0010] The aforementioned intermediation centrality specifically refers to the importance of a node in a network as an intermediary for the shortest path between other nodes. It is used to identify key hub nodes in an ecological network and quantifies the intermediation importance of a node by calculating the proportion of the number of shortest paths passing through that node to the total number of shortest paths in the network.

[0011] The clustering coefficient refers to the density of connections between neighboring nodes in the network, reflecting the local connectivity characteristics of the ecological network. It is measured by the ratio of the actual number of connections between neighboring nodes to the total number of possible connections between neighbors.

[0012] Specifically, the dynamic control limit refers to the range of early warning thresholds that are dynamically adjusted based on the statistical distribution characteristics of historical monitoring data. It can be adaptively updated with changes in time and environmental conditions. By establishing upper and lower control limits through statistical process control theory, it can achieve intelligent identification and early warning of abnormal changes in ecological indicators.

[0013] The monitoring frequency specifically refers to the time interval for data collection, sample analysis, and index calculation in the ecological restoration area. By adjusting the monitoring frequency, differentiated monitoring resource allocation can be achieved for areas with different ecological states. When the vegetation cover value of the salt marsh is in different ranges, a corresponding monitoring frequency adjustment strategy is adopted.

[0014] The sliding window technique specifically refers to a dynamic analysis method that uses a fixed-length time window to move across time-series data, updating statistical parameters and thresholds in real time. By sliding the time window, the timeliness of monitoring data is maintained, ensuring that early warning thresholds promptly reflect the latest changes in the ecosystem. The kriging interpolation and machine learning fusion algorithm combines traditional kriging spatial interpolation with machine learning techniques such as support vector regression. This hybrid algorithm improves spatial prediction accuracy by learning nonlinear relationships in historical monitoring data, expanding discrete monitoring point data into continuously distributed information across the entire region.

[0015] The network connectivity specifically refers to a quantitative indicator of the tightness of connections between nodes in an ecological network, reflecting the smooth flow of matter and energy within the ecosystem. It measures the overall connectivity level of the ecological network by calculating the ratio of the actual number of connected edges to the maximum possible number of connected edges.

[0016] The temporal graph convolutional network architecture is specifically a deep learning network structure based on multi-scale feature extraction. It includes an input layer for receiving multi-dimensional spatiotemporal ecological data, a graph convolutional layer for capturing spatial dependencies, a temporal convolutional layer for extracting time series features, a multi-head attention layer for adaptive weight allocation of key ecological succession features, and a fully connected output layer for generating predicted values ​​of ecological indicators for future time periods.

[0017] Specifically, the training dataset for the ecological succession trend prediction model is established by collecting historical monitoring data from the target area and similar marine ecological environments as the basic data source, performing quality checks and outlier removal on the original monitoring data, standardizing and preprocessing the data according to time series and spatial location, constructing a multi-dimensional feature vector containing the temporal changes of ecological indicators, spatial distribution characteristics, and environmental driving factors, dividing the data into training set, validation set, and test set, and establishing a data augmentation mechanism to expand the training sample size through sliding window sampling of time series data and rotation and flip transformation of spatial data.

[0018] This includes designing an ecological prediction adaptive adjustment function to adjust the multi-head attention weight allocation parameters of the ecological succession trend prediction model. The function calculates a comprehensive ecological status assessment value based on four data: network connectivity, vegetation cover, biodiversity index, and environmental stress intensity. When the comprehensive ecological status assessment value falls within different ranges, different weight adjustment strategies are used to adjust the weight allocation parameters of the multi-head attention mechanism. Through the dynamic weight adjustment mechanism, the adaptive prediction capability of the model for different ecological restoration stages is optimized.

[0019] This invention acquires comprehensive basic ecological parameters by constructing a multi-point monitoring network, uses a fusion algorithm of Kriging interpolation and machine learning to predict spatial continuous distribution, establishes a marine ecological network connectivity analysis model to quantify the intrinsic correlation of ecosystems, constructs a dynamic threshold ecological early warning algorithm system and a multi-parameter coupling analysis mechanism, and combines a time-series graph convolutional network-based ecological succession trend prediction model to form a complete spatiotemporal integrated ecological restoration effect evaluation technology system. This invention improves the spatial representativeness of monitoring data through spatial interpolation algorithms, enhances the temporal dimension evaluation capability through dynamic monitoring and prediction models, reveals the complex correlations between ecological elements through network analysis methods, and achieves comprehensive evaluation of different ecological indicators through a multi-parameter coupling mechanism, significantly improving the accuracy and reliability of ecological restoration effect evaluation. In summary, this invention solves the technical problem of inaccurate ecological restoration effect evaluation mentioned in the background art. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a graph used to verify the accuracy of the ecological succession trend prediction model in Example 2.

[0022] Figure 3 This is a trend chart of predicted vegetation cover and biodiversity index in Example 2.

[0023] Figure 4 This is a graph showing the temporal variation of ecological network connectivity in Example 2.

[0024] Figure 5 This is a panoramic view of the ecological restoration analysis and assessment in Example 2. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0026] like Figure 1 The diagram shown is a flowchart of a three-dimensional evaluation method for ecological restoration effects provided by this invention. This method includes the following steps:

[0027] S01. Deploy a multi-point monitoring network within the target restoration area, and use a combination of remote sensing images and field surveys to obtain basic ecological parameters such as species composition, restoration area, retained area, plant height, cover, density, survival rate and biomass of salt marsh vegetation. At the same time, record the three-dimensional spatial coordinates and timestamp information of each monitoring point.

[0028] S02. Based on the acquired discrete monitoring data, a fusion algorithm of Kriging interpolation and machine learning is used to predict the spatial continuous distribution of the study area. By analyzing the spatial location correlation characteristics, the influence range of each discrete point is calculated. The interpolation results are corrected using a support vector regression model to generate a distribution map of ecological indicators across the entire region.

[0029] S03. Establish a marine ecological network connectivity analysis model, abstract ecological patches as nodes and ecological corridors as edges, construct an ecological network graph, calculate topological indicators such as network connectivity, betweenness centrality, and clustering coefficient, and identify key ecological nodes through node importance ranking and network robustness analysis.

[0030] S04. Construct a dynamic threshold ecological early warning algorithm system, combine control chart theory to establish dynamic control limits for ecological indicators, use support vector machines to identify abnormal patterns, use sliding window technology to update threshold parameters, and automatically trigger an early warning mechanism when the monitored value exceeds the control limit.

[0031] S05. Establish a multi-parameter coupling analysis mechanism. When the vegetation cover value of the salt marsh is in the range of [60, 80], reduce the monitoring and processing frequency to 70% of the original frequency. When the cover value is in the range of (80, 95], maintain the current monitoring and processing frequency. When the cover value is in the range of (95, 100], increase the monitoring and processing frequency to 150% of the original frequency.

[0032] S06. Design an eigenvalue sensitivity analysis mechanism to calculate the sensitivity of the eigenvalues ​​of the ecological network adjacency matrix to changes in environmental stress parameters. Adjust the ecological restoration control strategy based on the sensitivity analysis results. When the eigenvalue change rate exceeds 15%, initiate emergency restoration measures.

[0033] S07. Construct an ecological succession trend prediction model, use a temporal graph convolutional network architecture to process spatiotemporal sequence data, use a multi-head attention mechanism to extract key ecological succession features, dynamically adjust the attention weight allocation according to three key parameters: network connectivity, vegetation cover and biodiversity index, predict the ecological restoration evolution trend in the next 6 months to 2 years and output it.

[0034] S08. Optionally, it also includes using 360-degree panoramic technology and time series comparative analysis methods to establish a three-dimensional display before and after restoration.

[0035] The monitoring and processing frequency refers to the time interval for data collection, sample analysis, and indicator calculation in the ecological restoration area. By adjusting the monitoring frequency, differentiated monitoring resource allocation can be achieved for areas with different ecological states.

[0036] The Kriging interpolation and machine learning fusion algorithm refers to a hybrid algorithm that combines traditional Kriging spatial interpolation methods with machine learning techniques such as support vector regression. By learning the nonlinear relationships in historical monitoring data, it improves the accuracy of spatial prediction.

[0037] The dynamic control limit refers to the range of early warning thresholds that are dynamically adjusted based on the statistical distribution characteristics of historical monitoring data, and can be adaptively updated as time and environmental conditions change.

[0038] Network connectivity refers to a quantitative indicator of the tightness of connections between nodes in an ecological network, reflecting the smooth flow of matter and energy within the ecosystem.

[0039] The term "intermediation centrality" refers to the importance of a node in a network as an intermediary for the shortest path between other nodes, and is used to identify key hub nodes in an ecological network.

[0040] The clustering coefficient refers to the density of interconnections between neighboring nodes in the network, reflecting the local connectivity characteristics of the ecological network.

[0041] The sliding window technique refers to a dynamic analysis method that uses a fixed-length time window to move across time series data and update statistical parameters and thresholds in real time.

[0042] The ecological network adjacency matrix refers to a two-dimensional matrix that describes the connection relationships between nodes in an ecological network. The matrix element values ​​represent the connection strength or connection status between nodes.

[0043] The specific structure of the ecological succession trend prediction model is a multi-scale feature extraction architecture based on a temporal graph convolutional network. It includes an input layer for receiving multi-dimensional spatiotemporal ecological data, a graph convolutional layer for capturing spatial dependencies, a temporal convolutional layer for extracting time series features, a multi-head attention layer for adaptive weight allocation of key ecological succession features, and a fully connected output layer for generating predicted values ​​of ecological indicators for future time periods. The weight allocation parameters of the multi-head attention mechanism are dynamically calculated based on three parameters: network connectivity, vegetation cover, and biodiversity index obtained from current monitoring. When network connectivity α∈[0.3, 0.6), the attention weights are tilted towards spatial features; when vegetation cover β∈[0.6, 0.9), the attention weights are tilted towards temporal features; and when the biodiversity index γ∈[2.0, 4.0], the attention weights are evenly distributed among spatiotemporal features.

[0044] The steps for establishing the training dataset for the ecological succession trend prediction model specifically include collecting historical monitoring data from the target area and similar marine ecological environments over the past 10 years as the basic data source; performing quality checks and outlier removal on the original monitoring data; standardizing and preprocessing the data according to time series and spatial location; constructing a multidimensional feature vector containing temporal changes in ecological indicators, spatial distribution characteristics, and environmental driving factors; dividing the data into training, validation, and test sets in a 7:2:1 ratio; and establishing a data augmentation mechanism to expand the training sample size through sliding window sampling of time series data and rotation and flipping transformation of spatial data.

[0045] The specific steps for training the ecological succession trend prediction model include: iteratively updating the model parameters using a gradient descent optimization algorithm; setting the learning rate to 0.001 and dynamically adjusting the learning rate using a cosine annealing scheduling strategy; using mean squared error as the loss function to measure the difference between the predicted and true values; introducing an L2 regularization term to prevent model overfitting; setting an early stopping mechanism to terminate training when the validation set loss has not improved for 10 consecutive cycles; using a mini-batch stochastic gradient descent method with a batch size of 32 to update the parameters; evaluating the model performance on the validation set and saving the optimal model weights every 5 cycles during training; and finally verifying the model's generalization ability and prediction accuracy through the test set. An adaptive adjustment function for ecological prediction is designed to adjust the weight allocation parameters of the multi-head attention mechanism in the ecological succession trend prediction model. The function calculates a comprehensive ecological status assessment value based on four data: network connectivity, vegetation cover, biodiversity index, and environmental stress intensity. Different weight adjustment strategies are adopted to adjust the weight allocation parameters of the multi-head attention mechanism when the comprehensive ecological status assessment value falls within different ranges. When the comprehensive ecological status assessment value δ∈[0, 0.4), the spatial feature attention weight is increased to 65% of the total weight. When δ∈[0.4, 0.7), the spatiotemporal feature attention weights are evenly distributed, each accounting for 50%. When δ∈[0.7, 1.0], the temporal feature attention weight is increased to 70% of the total weight. The adaptive prediction capability of the model for different ecological restoration stages is optimized through the dynamic weight adjustment mechanism.

[0046] The specific implementation methods of the above steps are described in detail below.

[0047] The specific implementation of step S01 involves first establishing a monitoring point network within the study area based on the geographical characteristics and ecological distribution patterns of the target restoration area using a systematic grid layout method. The spacing between monitoring points is set to 100m to 500m to ensure spatial coverage of over 85%. A Global Positioning System (GPS) device is installed at each monitoring point to record precise three-dimensional spatial coordinate information, with a coordinate accuracy requirement at the centimeter level. High-resolution remote sensing imagery is used to acquire macro-scale vegetation distribution information, with a spatial resolution of no less than 2m and a temporal resolution of once per month. Combined with field survey methods, detailed biological parameter measurements of the salt marsh vegetation are conducted, including determining vegetation cover using the quadrat method, measuring plant height using a ruler, determining density through counting, and determining biomass using the harvest method. The purpose of this step is to establish a comprehensive and accurate basic data collection system, providing high-quality raw data support for subsequent analysis.

[0048] The specific implementation of step S02 involves using the discrete monitoring data obtained in step S01 as input. First, a spatial correlation model is established using the Kriging interpolation method. By calculating the Euclidean distance and directional variability function between each monitoring point, the spatial autocorrelation range parameters are determined, typically set to 300m to 800m. Based on the Kriging interpolation, a support vector regression model is introduced for nonlinear correction. The kernel function of the support vector regression uses a radial basis function, with kernel parameters set between 0.1 and 1.0. The optimal parameter combination is determined through cross-validation. The machine learning model is trained using 70% of the historical monitoring data as the training set, 20% as the validation set, and 10% as the test set. The fusion algorithm combines the Kriging interpolation results with the support vector regression prediction results using a weighted average method, with the weight coefficients dynamically adjusted based on the prediction accuracy of the two methods. The purpose of this step is to overcome the limitations of traditional interpolation methods in handling complex nonlinear relationships and improve the accuracy and continuity of spatial prediction.

[0049] The specific implementation of step S03 is based on the ecological patch identification algorithm, which abstracts the ecological functional units within the study area into network nodes, and sets the ecological patch area threshold to... to Patches smaller than a threshold are merged. A minimum-cost path algorithm is used to identify ecological corridors, which are abstracted as network edges. The corridor width threshold is set to 10m to 50m. An adjacency matrix of the ecological network is constructed, with the matrix element values ​​representing connection strength using the reciprocal of the normalized distance. Network connectivity is calculated using connected component analysis in graph theory, traversing all nodes using a depth-first search algorithm to calculate the number of connected components. Betweenness centrality is calculated based on the shortest path algorithm, counting the frequency of each node acting as a mediator on the shortest path between other nodes. Clustering coefficient is calculated based on the triangle counting algorithm, counting the proportion of interconnected neighbors. This step quantifies the spatial structure characteristics and connectivity level of the ecosystem, identifying core nodes that play a crucial role in the entire ecological network.

[0050] The specific implementation of step S04 is to establish a dynamic control limit algorithm for ecological indicators based on control chart theory. First, the statistical distribution characteristics of historical monitoring data are calculated, including the mean, standard deviation, and coefficient of variation. The upper limit of the dynamic control limit is set to the mean plus three times the standard deviation, and the lower limit is set to the mean minus three times the standard deviation, with a confidence level of 99.7%. A support vector machine classifier is used to identify abnormal patterns. The training data includes normal state samples and abnormal state samples. Abnormal samples are obtained through manual annotation or expert knowledge. The sliding window technique uses a fixed-length time window of 30 time points. Statistical parameters and control limits are updated every time a new monitoring data point is added. A level one warning is triggered when the monitored value exceeds the control limit for three consecutive time points, and a level two warning is triggered when it exceeds it for five consecutive time points. The purpose of this step is to establish an adaptive ecological early warning mechanism that can promptly detect abnormal changes in the ecosystem and take corresponding management measures.

[0051] The specific implementation of step S05 involves establishing an adaptive monitoring frequency adjustment mechanism based on the vegetation cover of the salt marsh. The monitoring frequency is determined by calculating the vegetation cover value at each monitoring point in real time. When the cover value is between 60% and 80%, the ecological state is relatively stable, and the monitoring frequency is reduced to 70% of the original frequency, i.e., monitoring is changed from weekly to once every 10 days. When the cover value is between 80% and 95%, the ecological state is in dynamic equilibrium, and the current monitoring frequency is maintained. When the cover value is between 95% and 100%, the ecological state is approaching saturation and may change, so the monitoring frequency is increased to 150% of the original frequency, i.e., monitoring is changed from weekly to once every 5 days. The frequency adjustment uses an exponential smoothing method for buffering to avoid resource waste caused by frequent changes. The purpose of this step is to optimize the allocation of monitoring resources, improving monitoring efficiency and economy while ensuring monitoring quality.

[0052] The specific implementation of step S06 is based on sensitivity analysis using the eigenvalue decomposition theory of the ecological network adjacency matrix. First, the maximum eigenvalue of the ecological network adjacency matrix is ​​calculated, reflecting the overall connectivity level of the network. Environmental stress parameters include water pollution index, climate change index, and anthropogenic disturbance intensity, each represented by a normalized value between 0 and 1. Sensitivity calculation employs numerical differentiation, observing the degree of change in eigenvalues ​​through small perturbations to each environmental stress parameter. When a 1% change in an environmental stress parameter leads to a change in eigenvalue exceeding 0.15%, that parameter is considered highly sensitive to the ecological network. The rate of change of eigenvalues ​​is calculated using relative change, i.e., the difference between the new and original eigenvalues ​​divided by the original eigenvalue. When the cumulative rate of change of eigenvalues ​​exceeds 15%, emergency remediation measures are automatically initiated, including increasing human intervention and adjusting remediation strategies. The purpose of this step is to quantitatively assess the impact of environmental factors on the stability of the ecological network, providing a scientific basis for ecological restoration management.

[0053] The specific implementation of step S07 involves constructing an ecological succession trend prediction model based on a temporal graph convolutional network. This model employs a multi-layered architecture to process spatiotemporal sequence data. The input layer receives spatiotemporal data containing multi-dimensional ecological indicators such as vegetation cover, biomass, and species diversity. Data preprocessing uses a minimum-maximum normalization method to normalize all indicators to the range of 0 to 1. The graph convolutional layer uses a spectral domain graph convolution algorithm to capture spatial dependencies, with a kernel size of 3×3 and a feature map channel count of 64 to 256. The temporal convolutional layer uses a one-dimensional convolutional neural network to extract time-series features, with a kernel length of 3 to 7 time steps and a stride of 1. The multi-head attention layer contains 8 attention heads, each with a dimension of 64, and calculates the association weights between different ecological features through a self-attention mechanism. The weighting strategy dynamically adjusts based on three parameters: network connectivity α, vegetation cover β, and biodiversity index γ. When α ∈ [0.3, 0.6), spatial features account for 65% of the weight, and temporal features account for 35%. When β ∈ [0.6, 0.9), temporal features account for 60%, and spatial features account for 40%. When γ ∈ [2.0, 4.0], both spatiotemporal features account for 50%. The fully connected output layer uses linear transformation to generate predicted ecological indicators for the next 6 months to 2 years. This step aims to achieve quantitative prediction of ecosystem succession trends, providing forward-looking guidance for ecological restoration planning and management.

[0054] Step S08 is optional. Its specific implementation involves establishing a three-dimensional display system for pre- and post-restoration data using a 360-degree panoramic camera and drone aerial photography equipment. The panoramic camera resolution is no less than 4K, and the shooting interval is set to once a month. Time-series comparative analysis employs image registration algorithms to ensure spatial consistency of images from different periods, with registration accuracy required to reach the sub-pixel level. The three-dimensional display system constructs a 3D visualization environment based on virtual reality technology, allowing users to interactively view the ecological restoration effects from different periods and perspectives. Data fusion analysis compares and verifies the output results of the prediction model with measured data, calculating prediction accuracy indicators including root mean square error, mean absolute error, and coefficient of determination. A comprehensive evaluation report is automatically generated, including restoration effect evaluation, trend prediction, and risk warning. The report format uses a standardized template to ensure the accuracy and consistency of information transmission. The long-term tracking and monitoring database is constructed using a relational database management system, supporting the storage, query, and analysis of massive amounts of spatiotemporal data. The purpose of this step is to establish an intuitive and visual evaluation and display platform, providing comprehensive information support for ecological restoration effectiveness evaluation and decision-making.

[0055] The detailed structure of the ecological succession trend prediction model comprises five main components. First, the input layer receives multidimensional spatiotemporal ecological data in a four-dimensional tensor format, with dimensions corresponding to the number of samples, time step, number of spatial nodes, and number of features. The graph convolutional layer employs a Chebyshev multinomial approximation spectral domain graph convolution method, effectively extracting spatial information through eigenvalue decomposition of the Laplacian matrix. This layer contains 3 to 5 graph convolutional units, each followed by batch normalization and an activation function. The temporal convolutional layer uses causal convolution to ensure the causal relationship of the time series, with 2 to 4 layers, each containing residual connections and dropout layers to prevent overfitting. The multi-head attention layer calculates attention scores using three linear transformation matrices: query, key, and value. These attention scores are normalized using a soft maximum function and multiplied by the numerical vector to obtain a weighted feature representation. The fully connected output layer contains 2 to 3 fully connected sub-layers. The output dimension of the last layer equals the number of predicted targets, and a linear activation function ensures the continuity of the predicted values.

[0056] The detailed steps for establishing the training dataset for the ecological succession trend prediction model are as follows: First, historical monitoring data from the target area and similar marine ecological environments over the past 10 years were collected. Data sources included satellite remote sensing data, field survey data, and environmental monitoring station data. Data quality control employed statistical methods to identify outliers, including outlier detection based on interquartile range and outlier discrimination based on standard deviation. The outlier removal rate was controlled below 5% to avoid excessive cleaning. Data preprocessing included time-series interpolation to fill missing values, spatial coordinate system standardization, and spatiotemporal registration of multi-source data. Cubic spline interpolation was used to maintain data smoothness. Feature vector construction integrated the temporal changes, spatial distribution characteristics, and environmental driving factors of ecological indicators into a unified multi-dimensional vector, typically with 50 to 200 dimensions. Data was strictly partitioned according to chronological order, with the training set comprising 70%, the validation set 20%, and the test set 10%, ensuring the objectivity of model evaluation. The data augmentation mechanism increases the number of training samples by using sliding window sampling. The window length is set to 2 to 3 times the prediction time, and the sliding step size is set to 1 time unit. At the same time, geometric transformations such as rotation, flipping, and scaling of spatial data are used to expand the diversity of samples. The size of the training set after data augmentation increases to 3 to 5 times the original size.

[0057] It should be noted that the key technical ideas of this invention are mainly reflected in three aspects. First, the spatial prediction technology that integrates Kriging interpolation and machine learning breaks through the limitation of traditional interpolation methods that rely solely on spatial correlation. By introducing a support vector regression model to learn complex nonlinear relationships in historical data, it significantly improves the accuracy and robustness of spatial continuous distribution prediction. Compared with traditional Kriging methods, it can better handle the heterogeneity and abrupt changes in ecosystems, providing a more reliable data foundation for the accurate quantification of ecological restoration effects. Second, the graph theory-based ecological network connectivity analysis method transforms complex ecological spatial relationships into a computable network topology. By analyzing node importance and network robustness, it identifies key ecological nodes. Compared with traditional single-point assessment methods, it can grasp the overall effect and key links of ecological restoration from a system level, providing systematic scientific guidance for the formulation of ecological protection and restoration strategies. Third, the dynamic threshold early warning mechanism, combined with sliding window technology, achieves adaptive updates of early warning parameters, overcoming the shortcomings of fixed thresholds that cannot adapt to environmental changes. By using support vector machine anomaly pattern recognition, it improves the accuracy and timeliness of early warning. Compared with traditional static threshold methods, it can more effectively capture the dynamic change characteristics of ecosystems.

[0058] The synergistic effect of these three key technological approaches forms a complete three-dimensional evaluation system. Spatial prediction technology provides high-quality continuous data support for network analysis; key nodes identified by network analysis provide key monitoring targets for the early warning system; and the dynamic feedback from the early warning system provides real-time information for adjusting the parameters of the spatial prediction model. These three elements mutually promote each other, forming a closed-loop optimization mechanism. Compared to existing single-technology paths or simple combinations, this synergistic system can simultaneously consider accuracy, systematicity, and adaptability, achieving a fundamental shift in ecological restoration effect assessment from qualitative description to quantitative analysis, from static evaluation to dynamic monitoring, and from local observation to systemic understanding. This provides a scientifically sound and comprehensive technical support system for marine ecological restoration.

[0059] It should be noted that this invention also solves the following technical problem: Existing technologies often lack intelligent adaptive adjustment capabilities in ecological restoration monitoring. Traditional ecological restoration monitoring systems mostly employ fixed-frequency monitoring modes, failing to dynamically adjust according to different stages of ecological restoration and changing ecological conditions. This leads to problems such as unreasonable allocation of monitoring resources, low monitoring efficiency, and poor cost control. This invention establishes a multi-parameter coupling analysis mechanism to automatically adjust the monitoring frequency based on different ranges of vegetation cover values ​​in salt marshes. When the cover value falls within different ranges, differentiated monitoring strategies are adopted, achieving intelligent allocation and adaptive adjustment of monitoring resources, thus improving the operational efficiency and economy of the monitoring system. Simultaneously, this invention designs an eigenvalue sensitivity analysis mechanism to calculate the sensitivity of the ecological network adjacency matrix eigenvalues ​​to changes in environmental stress parameters. Based on the sensitivity analysis results, the ecological restoration control strategy is automatically adjusted. When the eigenvalue change rate exceeds a preset threshold, emergency restoration measures are initiated, achieving intelligent adjustment of the restoration strategy and risk prevention and control, effectively overcoming the technical limitations of traditional technologies that lack adaptive adjustment capabilities. Furthermore, this invention addresses the technical problem of the lack of long-term succession trend prediction capabilities for ecological restoration effects. Traditional assessment methods are mainly based on the current state or short-term changes, failing to predict the long-term succession direction and development trend of ecosystems, thus making it difficult to provide a scientific basis for long-term restoration planning. This invention constructs an ecological succession trend prediction model based on a time-series graph convolutional network and utilizes a multi-head attention mechanism to adaptively extract key ecological succession features, enabling the prediction of ecological restoration evolution trends over the next 6 months to 2 years. This provides important technical support for long-term planning and management decisions in ecological restoration.

[0060] Specifically, the principle of this invention is as follows: The core technical problem of inaccurate ecological restoration effect assessment lies in the establishment of a multi-dimensional data fusion analysis system and an intelligent, precise assessment mechanism. At the data acquisition level, the deployment of a multi-point monitoring network significantly increases the spatial density and temporal frequency of monitoring data, providing a sufficient data foundation for accurate assessment and overcoming the assessment bias caused by data sparsity in traditional methods. At the spatial analysis level, the fusion algorithm of Kriging interpolation and machine learning expands discrete monitoring point data into continuous distribution information across the entire domain by learning the nonlinear relationships in historical monitoring data. The support vector regression model can capture complex spatial variation patterns, effectively improving spatial prediction accuracy and solving the problem that traditional interpolation methods cannot accurately reflect the spatial distribution characteristics of ecological indicators. At the system correlation analysis level, the marine ecological network connectivity analysis model establishes a topological description system of the ecosystem by abstracting ecological patches and ecological corridors into network nodes and edges. Topological indicators such as network connectivity, betweenness centrality, and clustering coefficient can quantitatively characterize the interaction relationships between ecological elements, revealing the internal organizational structure and functional characteristics of the ecosystem and compensating for the lack of systematic analysis in traditional methods. At the dynamic monitoring level, the dynamic threshold ecological early warning algorithm system, combined with the dynamic control limits established by control chart theory, can adaptively update with changes in time and environmental conditions. Support vector machine anomaly pattern recognition technology improves the accuracy of anomaly detection, and sliding window technology ensures real-time updates of threshold parameters, achieving precise dynamic monitoring of the ecological restoration process. At the parameter coupling analysis level, the multi-parameter coupling analysis mechanism and eigenvalue sensitivity analysis mechanism, by comprehensively considering the interaction of multiple ecological indicators and the effects of environmental stress factors, establish a multi-dimensional parameter coupling evaluation system, avoiding the one-sidedness of single-indicator evaluation. At the predictive evaluation level, the temporal graph convolutional network architecture can simultaneously process time series and spatial relationship information. The multi-head attention mechanism dynamically adjusts the weight allocation based on network connectivity, vegetation cover, and biodiversity index, achieving accurate prediction of ecological succession trends and providing forward-looking support for the evaluation results.

[0061] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0062] The specific implementation of step S01 involves first establishing a monitoring point network within the study area based on the geographical characteristics and ecological distribution patterns of the target restoration area using a systematic grid deployment method. The spacing between monitoring points is set to 100m to 500m to ensure spatial coverage of over 85%. A Global Positioning System (GPS) device is installed at each monitoring point to record precise three-dimensional spatial coordinate information, with a required coordinate accuracy at the centimeter level. The three-dimensional spatial coordinates are represented as follows: In the formula, For the first Spatial coordinates of each monitoring point , For planar coordinates, Elevation coordinates , This represents the total number of monitoring points. High-resolution remote sensing imagery was used to acquire macro-scale vegetation distribution information, with a spatial resolution of at least 2 meters and a temporal resolution of once per month. Detailed biological parameters of the salt marsh vegetation were measured using field survey methods, including determining vegetation cover using the quadrat method, measuring plant height using a ruler, determining density through counting, and determining biomass using the harvest method.

[0063] The specific implementation of step S02 involves using the discrete monitoring data obtained in step S01 as input. First, a spatial correlation model is established using the Kriging interpolation method. By calculating the Euclidean distance and directional variability function between each monitoring point, the spatial autocorrelation range parameter is determined, typically set to 300m to 800m. The weight calculation formula for Kriging interpolation is as follows: In the formula, Spatial location to be predicted The interpolation result at the location, Two-dimensional spatial coordinates , For the first Spatial location of each monitoring point The observed values, Two-dimensional spatial coordinates of the monitoring points , For the first The weighting coefficient of each monitoring point. This represents the number of monitoring points participating in Kriging interpolation. The weighting coefficients are calculated based on the Kriging equations. In the formula, For the summation index, , For monitoring points and The semivariogram values ​​between Let be the Lagrange multiplier. The formula for calculating the semivariogram is: In the formula, For spatial distance variables, Distance The semivariogram value at that location, Distance The data at the location is of quantity. For indexing data pairs, , and Distance The observations at two spatial locations, among which Indicates distance and location for Another spatial location. Based on Kriging interpolation, a support vector regression model is introduced for nonlinear correction. The prediction function of support vector regression is... In the formula, For the predicted output of support vector regression, The input feature vector contains spatial coordinates and environmental factors. For the first Feature vectors of support vectors , and It is a Lagrange multiplier. For radial basis kernel functions, For bias terms, This represents the number of support vectors. The radial basis function kernel function expression is: In the formula, The radial basis function (RBF) kernel parameter is set from 0.1 to 1.0. The Euclidean distance between the feature vectors is given. The final prediction result of the fusion algorithm is... In the formula, The fused predicted value, For the Kriging interpolation results, For support vector regression prediction results, and For the weighting coefficients, satisfying .

[0064] The specific implementation of step S03 is based on the ecological patch identification algorithm, which abstracts the ecological functional units within the study area into network nodes, and sets the ecological patch area threshold to... to Patches smaller than a threshold are merged. An adjacency matrix of the ecological network is constructed. The matrix element values ​​represent the connection strength using the normalized reciprocal of the distance. In the formula, For nodes With nodes The strength of the connection between them , For nodes With nodes The Euclidean distance between them This represents the maximum distance in the network. The formula for calculating network connectivity is: In the formula, For network connectivity, Let be the total number of nodes in the ecological network. The formula for calculating betweenness centrality is: In the formula, For nodes The centrality of the middle, and Let be any two distinct nodes in the network. and , For nodes To the node The total number of shortest paths, For the nodes The number of shortest paths. The formula for calculating the clustering coefficient is: In the formula, For nodes The clustering coefficient, For nodes The actual number of connections between neighboring nodes. For nodes The degree.

[0065] The specific implementation of step S04 is to establish a dynamic control limit algorithm for ecological indicators based on control chart theory. First, the statistical distribution characteristics of historical monitoring data are calculated, including the mean, standard deviation, and coefficient of variation. The formula for calculating the dynamic control limit is as follows: and In the formula, For the first Upper control limit of time, For the first Lower control limit at time, For the first The average of the data at time points. For the first The standard deviation of the data at time points. The formulas for calculating the mean and standard deviation within the sliding window are as follows: and In the formula, The sliding window length is set to 30 time points. For time indexing, , For the first The monitoring values ​​at each time point. The anomaly detection function is... In the formula, For the first The anomaly detection result at any given time: 1 indicates an anomaly, and 0 indicates normal.

[0066] The specific implementation of step S05 is to establish an adaptive monitoring frequency adjustment mechanism based on the vegetation cover of the salt marsh, wherein the monitoring frequency adjustment function is: In the formula, The adjusted monitoring frequency, This is the original monitoring frequency. This is the frequency adjustment coefficient. This represents the vegetation cover value. The formula for calculating the frequency adjustment coefficient is: In the formula, the frequency adjustment coefficient is set in segments according to different ranges of vegetation cover value.

[0067] The specific implementation of step S06 is based on sensitivity analysis using the eigenvalue decomposition theory of the ecological network adjacency matrix. First, the ecological network adjacency matrix is ​​calculated. Maximum eigenvalue The formula for calculating the rate of change of eigenvalues ​​is: In the formula, The rate of change of the eigenvalue. The maximum eigenvalue after perturbation. This represents the maximum eigenvalue before the disturbance. The sensitivity calculation formula for environmental stress parameters is as follows: In the formula, For the first Sensitivity of environmental stress parameters These correspond to the water pollution index, climate change index, and intensity of human disturbance, respectively. For the first Several environmental stress parameters, including water pollution index, climate change index, and anthropogenic disturbance intensity. Emergency remediation trigger conditions are... .

[0068] The specific implementation of step S07 involves constructing an ecological succession trend prediction model based on a temporal graph convolutional network. The weight allocation function of the multi-head attention mechanism is based on the network connectivity. Vegetation coverage Biodiversity Index Dynamic adjustments are made. The weight allocation function is: In the formula, The attention weights are for spatial features. The attention weights are for temporal features. The formula for calculating the comprehensive ecological status assessment value is as follows: In the formula, This is a comprehensive assessment value of the ecological status. The intensity of environmental stress is a comprehensive indicator that includes water pollution, climate change, and anthropogenic disturbances. For the corresponding weight coefficients, satisfying The ecological prediction adaptive adjustment function is: .

[0069] The specific implementation method of step S08 is the same as described above, and will not be repeated in detail here.

[0070] It should be explained that the Kriging interpolation and machine learning fusion algorithm overcomes the limitations of a single interpolation method by combining traditional spatial interpolation methods with a nonlinear regression model. The weight coefficients in the fusion algorithm are dynamically adjusted based on the prediction accuracy of the two methods, and the weight optimization function is: , In the formula, The root mean square error of Kriging interpolation. The root mean square error of support vector regression;

[0071] When the root mean square error of cross-validation for Kriging interpolation is small Larger values, and vice versa With larger values, this adaptive weight allocation mechanism significantly improves the accuracy and stability of spatial prediction, increasing prediction accuracy by 15% to 25% compared to traditional Kriging interpolation. In the ecological network connectivity analysis model, the adjacency matrix represents connection strength using the reciprocal of the normalized distance, and the connection strength decay function is... ,in The characteristic distance parameter represents the decrease in connection strength to [value missing]. The characteristic distance at time;

[0072] This system effectively quantifies the spatial relationships between ecological patches. Network connectivity indicators reflect the overall connectivity level of the ecosystem, betweenness centrality identifies key hub nodes in the ecological network, and clustering coefficients describe the density of local ecological networks. This comprehensive network analysis indicator system provides a scientific basis for optimizing the spatial layout of ecological restoration. The dynamic threshold ecological early warning algorithm system uses sliding window technology to achieve adaptive updating of the early warning threshold. The dynamic threshold adjustment function is... ,in For the first The adaptive adjustment factor at time point ranges from 0.1 to 0.5.

[0073] Compared to traditional fixed threshold methods, dynamic control limits can adapt to the natural changing trends of ecosystems, effectively reducing the false positive warning rate and improving the accuracy and practicality of warnings. The eigenvalue sensitivity analysis mechanism calculates the response of the ecological network's adjacency matrix eigenvalues ​​to changes in environmental stress parameters; the sensitivity matrix calculation function is... ,in For the sensitivity matrix, Let be the environmental stress parameter vector. For ecological network adjacency matrix, and These are the corresponding left and right feature vectors;

[0074] The impact of different environmental factors on the stability of the ecological network was quantitatively assessed, providing a quantitative basis for determining the key directions of ecological restoration management. Emergency restoration measures are automatically initiated when the rate of change of characteristic values ​​exceeds a threshold of 15%, enabling early identification and timely response to ecological risks. The multi-head attention weight allocation function in the ecological succession trend prediction model is dynamically adjusted based on three key parameters of the current ecological state. The attention weight normalization function is as follows: ,in The final attention weights after normalization. These are the original attention weight values;

[0075] This adaptive weighting mechanism enables the model to automatically adjust its focus on spatial and temporal features according to different stages of ecological restoration, significantly improving the adaptability and accuracy of the prediction model. Compared with the fixed-weight prediction model, the average prediction error is reduced by 20% to 30%.

[0076] To better understand and implement this invention, a specific application scenario is provided as Example 2: A technical team used the three-dimensional evaluation method of this invention to conduct a 24-month monitoring and evaluation of the ecological restoration effect of a nearshore salt marsh wetland ecological restoration project in a certain sea area. The restoration area was 186.7 square kilometers. The main restoration targets are typical salt marsh plant communities such as Suaeda salsa, Reed and Ephedra sinica.

[0077] The technical team first established a multi-site monitoring network consisting of 32 monitoring points within the target restoration area, using a combination of Sentinel-2 multispectral remote sensing imagery and field surveys to obtain basic ecological parameters. The remote sensing imagery had a spatial resolution of 10m and a temporal resolution of 5 days. The field survey employed the quadrat method, with three 5m × 5m quadrats set up at each monitoring point. Through 18 months of continuous monitoring, basic ecological parameters such as species composition, restoration area, retained area, plant height, canopy cover, density, survival rate, and biomass of the salt marsh vegetation were obtained. These are shown in Table 1.

[0078] Table 1. Statistical Table of Basic Ecological Parameters of Salt Marsh Vegetation

[0079]

[0080] During the monitoring period, a total of 7,296 sets of three-dimensional spatial coordinate data were recorded, with a coordinate accuracy of ±0.5m and a timestamp accuracy at the second level. Based on the acquired discrete monitoring data, the technical team used a fusion algorithm of Kriging interpolation and machine learning to predict the spatial continuous distribution of the study area. The prediction accuracy of the traditional Kriging interpolation method was 76.3%, which was improved to 89.7% after being corrected by fusing a support vector regression model. By analyzing the spatial location correlation characteristics, the algorithm calculated the influence range of each discrete point from 65 to 132m, generating a global ecological indicator distribution map covering indicators such as vegetation cover, biomass density, and species diversity.

[0081] The technical team established a marine ecological network connectivity analysis model, abstracting 186 ecological patches as nodes and 423 ecological corridors as edges, constructing an ecological network graph containing 609 connections. The calculated network connectivity was 0.47, indicating good overall network connectivity. Through betweenness centrality calculations, 15 key ecological nodes were identified, with betweenness centrality values ​​ranging from 0.23 to 0.78. Clustering coefficient analysis showed an average clustering coefficient of 0.34, indicating good local connectivity characteristics of the ecological network. Node importance ranking results showed that three large reed patches located at the river estuary had the highest importance, playing a crucial role in maintaining the stability of the entire network. Network robustness analysis showed that when the top 10% of important nodes were removed, the network connectivity only decreased to 0.39, indicating strong anti-interference capabilities of the network.

[0082] Regarding the construction of a dynamic threshold ecological early warning algorithm system, the technical team established dynamic control limits for ecological indicators by combining control chart theory. Taking vegetation cover as an example, based on historical monitoring data, the dynamic control upper limit was calculated to be 85.6%, and the dynamic control lower limit was 52.3%. Support vector machine algorithms were used to identify abnormal patterns, achieving an anomaly detection accuracy of 92.8%. A 30-day sliding window technique was used to update threshold parameters, with the sliding window updating every 5 days. During the monitoring period, the early warning mechanism was triggered 47 times, of which 36 were valid warnings, resulting in an early warning accuracy of 76.6%. When monitored values ​​exceeded the control limits, the system automatically generated an early warning report and notified relevant personnel via SMS and email.

[0083] The established multi-parameter coupled analysis mechanism adjusts the monitoring and processing frequency based on different ranges of vegetation cover values ​​in the salt marsh. When the cover value is in the range of 60–80, the monitoring and processing frequency is reduced from twice a week to 1.4 times a week, saving 30% of monitoring resources. When the cover value is in the range of 80–95, the monitoring and processing frequency is maintained at twice a week. When the cover value is in the range of 95–100, the monitoring and processing frequency is increased to three times a week, ensuring precise monitoring of high-quality ecological areas. After 18 months of implementation, this mechanism improved overall monitoring efficiency by 23.5% while ensuring monitoring quality.

[0084] The eigenvalue sensitivity analysis mechanism calculated the sensitivity of the eigenvalues ​​of the ecological network adjacency matrix to changes in environmental stress parameters. The results showed that temperature change had the highest sensitivity to network eigenvalues, with a sensitivity coefficient of 2.34, followed by salinity change, with a sensitivity coefficient of 1.87. pH change had a sensitivity coefficient of 1.52, and nutrient concentration change had a sensitivity coefficient of 1.23. During the monitoring period, there were eight instances where the eigenvalue change rate exceeded 15%, and emergency remediation measures were promptly initiated in all cases, including replanting, water quality regulation, and soil improvement, effectively preventing further degradation of the ecosystem.

[0085] The constructed ecological succession trend prediction model employs a temporal graph convolutional network architecture to process spatiotemporal sequence data. The model consists of 4 graph convolutional layers, 3 temporal convolutional layers, and an 8-head attention mechanism. The training dataset contains nearly 10 years of historical monitoring data, totaling 23,456 records, divided into training, validation, and test sets in a 7:2:1 ratio. The model was trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 32, converging after 150 epochs. Figure 2 As shown, the root mean square error of the model on the validation set is 0.087, and the coefficient of determination is... The value of 0.923 indicates that the model has high prediction accuracy.

[0086] The weight allocation of the multi-headed attention mechanism was dynamically adjusted based on three key parameters: network connectivity, vegetation cover, and biodiversity index. During the monitoring period, network connectivity α ranged from 0.35 to 0.58. When α ∈ 0.3–0.6, the attention weight was biased towards spatial features, accounting for 65%. Vegetation cover β ranged from 0.62 to 0.89. When β ∈ 0.6–0.9, the attention weight was biased towards temporal features, accounting for 70%. Biodiversity index γ ranged from 2.3 to 3.7. When γ ∈ 2.0–4.0, the attention weight was evenly distributed between spatiotemporal features, each accounting for 50%.

[0087] The ecological prediction adaptive adjustment function calculates the comprehensive ecological state assessment value δ based on four key parameters, as shown in Table 2.

[0088] Table 2 Statistical Table of Comprehensive Ecological Status Assessment Values

[0089]

[0090] Based on changes in the comprehensive evaluation value, the system dynamically adjusts the multi-head attention weight allocation. In the initial three months of the project, the δ value was 0.52, falling within the range of 0.4 to 0.7, and the system evenly distributed spatiotemporal feature attention weights, each accounting for 50%. As the restoration progressed, the δ value reached 0.74 at 18 months, falling within the range of 0.7 to 1.0, and the system increased the temporal feature attention weight to 70%, paying greater attention to the temporal patterns of ecological succession.

[0091] The model successfully predicted the ecological restoration evolution trend over the next 6 months to 2 years. For example... Figure 3 As shown, the prediction results indicate that vegetation cover will steadily increase to 83.5% over the next 12 months, the biodiversity index will reach 3.8, and network connectivity will increase to 0.52. Prediction accuracy verification shows that the average absolute percentage error (APE) for the 6-month prediction is 8.3%, for the 12-month prediction is 13.7%, and for the 24-month prediction is 21.2%.

[0092] The technical team also used 360-degree panoramic technology to create a three-dimensional display system showcasing the restoration process before and after. For example... Figure 4 As shown, the system integrates 12 high-resolution panoramic cameras, capturing 4K resolution images at 30fps, enabling 360-degree panoramic recording without blind spots. A three-dimensional comparison display function before and after restoration was established using time-series comparative analysis. Figure 5 As shown, it displays marine environmental data maps and a panoramic view of ecological restoration analysis and assessment.

[0093] The system integrates the output of the ecological succession trend prediction model with measured data to generate a comprehensive assessment report. It also establishes a long-term monitoring database, storing over 450,000 monitoring data records accumulated over 18 months, providing crucial data support for subsequent ecological restoration and management. The database employs a distributed storage architecture, supporting concurrent access by multiple users, with query response time controlled within 500ms.

[0094] Compared to traditional ecological restoration assessment methods, this invention utilizes a multi-point monitoring network and remote sensing image fusion technology to shift from point-based monitoring to area-based monitoring, significantly improving the spatial coverage and temporal continuity of data acquisition. The Kriging interpolation and machine learning fusion algorithm overcomes the linear assumptions of traditional spatial interpolation methods, capturing complex nonlinear spatial relationships and significantly improving spatial prediction accuracy. The ecological network connectivity analysis model analyzes the ecological restoration effect from a systems theory perspective, overcoming the limitations of traditional methods that focus only on a single ecological indicator, and providing a scientific basis for the holistic assessment of the ecosystem. The dynamic threshold ecological early warning algorithm system breaks free from the constraints of fixed thresholds, adaptively adjusting early warning parameters according to the evolutionary laws of the ecosystem, significantly improving the accuracy and timeliness of early warnings. The multi-parameter coupling analysis mechanism enables precise allocation of monitoring resources, avoiding the resource waste caused by traditional uniform monitoring. The ecological succession trend prediction model based on time-series graph convolutional networks integrates the advantages of time series analysis and graph neural networks, simultaneously capturing the temporal evolution laws and spatial correlation characteristics of the ecosystem, providing scientific prediction support for long-term planning and management of ecological restoration. 360-degree panoramic technology and three-dimensional display system enable the visualization of ecological restoration effects, providing decision-makers with an intuitive platform for displaying evaluation results.

[0095] It should be noted that the variables involved in this invention are explained in detail in Tables 3 and 4.

[0096] Table 3. Variable Explanation Table (Part 1)

[0097]

[0098] Table 4. Variable Explanation Table (Part Two)

[0099]

[0100] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the three-dimensional effect of ecological restoration, characterized in that, A multi-point monitoring network is laid out in the target restoration area, and remote sensing images and field investigation are combined to obtain the basic ecological parameters of the salt marsh vegetation, while recording the three-dimensional spatial coordinates and timestamp information of each monitoring point; based on the obtained discrete monitoring data, a Kriging interpolation and machine learning fusion algorithm is used to predict the spatial continuous distribution of the study area, the influence range of each discrete point is calculated by analyzing the spatial position correlation characteristics, the support vector regression model is used to correct the interpolation results, and the global ecological index distribution map is generated; an ocean ecological network connectivity analysis model is established, the ecological patches are abstracted as nodes, the ecological corridors are abstracted as edges, an ecological network graph is constructed, topological indexes are calculated, key ecological nodes are identified through node importance sorting and network robustness analysis; a dynamic threshold ecological early warning algorithm system is constructed, a dynamic control limit of the ecological index is established combining with the control chart theory, an abnormal pattern is identified using a support vector machine, and a sliding window technology is used to update the threshold parameters, and the early warning mechanism is automatically triggered when the monitoring value exceeds the control limit; a multi-parameter coupling analysis mechanism is established, the monitoring processing frequency is adjusted according to the salt marsh vegetation coverage value; a feature value sensitivity analysis mechanism is designed, the sensitivity of the eigenvalues of the ecological network adjacency matrix to the change of the environmental stress parameters is calculated, and the ecological restoration control strategy is adjusted according to the sensitivity analysis result; An ecological succession trend prediction model is constructed, a time series graph convolution network architecture is used to process the spatio-temporal sequence data, a multi-head attention mechanism is used to extract key ecological succession features, the attention weight distribution is dynamically adjusted according to the network connectivity, vegetation coverage and biodiversity index, the future ecological restoration evolution trend is predicted and output; The topological indexes include network connectivity, intermediate centrality and clustering coefficient, the intermediate centrality specifically refers to the importance of a node in the network as an intermediate in the shortest path between other nodes, and is used to identify key hub nodes in the ecological network, and the intermediate importance of the node is quantified by calculating the proportion of the number of shortest paths passing through the node in the total number of shortest paths in the network; the dynamic control limit specifically refers to the early warning threshold range dynamically adjusted based on the statistical distribution characteristics of the historical monitoring data; the environmental stress parameters include water pollution index, climate change index and human disturbance intensity; The multi-parameter coupling analysis mechanism is established, and the monitoring processing frequency is adjusted according to the salt marsh vegetation coverage value, specifically: when the salt marsh vegetation coverage value is in the interval [60, 80], the monitoring processing frequency is reduced to 70% of the original frequency, when the coverage value is in the interval (80, 95], the current monitoring processing frequency is maintained, and when the coverage value is in the interval (95, 100], the monitoring processing frequency is increased to 150% of the original frequency.

2. The method of claim 1, wherein the method is used for the three-dimensional evaluation of the ecological restoration effect. The step of laying out the multi-point monitoring network specifically establishes a spatially uniformly distributed monitoring point grid in the target restoration area, combines remote sensing image technology to obtain macro ecological condition information, and collects detailed vegetation ecological parameters through field investigation, to ensure the spatial representativeness and time continuity of the monitoring data, and to provide basic data support for subsequent spatial interpolation analysis and ecological network construction.

3. The method of claim 2, wherein the method further comprises: The base ecological parameter, in particular, is a quantitative ecological index obtained by monitoring, while recording the three-dimensional spatial coordinates and timestamp information of each monitoring point, forming a complete spatio-temporal ecological data set, providing a data basis for quantitative evaluation of ecological restoration effect.

4. The method of claim 3, wherein the method further comprises: The ecological network adjacency matrix, in particular, is a two-dimensional matrix describing the connection relationship between nodes in the ecological network, and the matrix element value represents the connection strength or connection state between nodes. The stability of the ecological network structure and the influence of environmental stress on the ecological system are reflected by the change of the matrix eigenvalue.

5. The method for assessing the effect of ecological restoration in three dimensions according to claim 4, characterized in that, Before constructing the ecological network graph, the preprocessing step of accurately identifying the boundary of ecological patch and assessing the connectivity of ecological corridor is also included. The spatial analysis technique is used to determine the geometric shape and area size of the ecological patch, and to identify the width and connectivity of the ecological corridor, providing accurate spatial basic information for the abstract modeling of nodes and edges of the ecological network.

6. The method of claim 5, wherein the method further comprises: The weight distribution parameter of the multi-head attention mechanism is dynamically calculated according to the network connectivity, vegetation cover and biodiversity index obtained by current monitoring. When the network connectivity is in different intervals, the attention weight tilts towards the spatial feature; when the vegetation cover is in different intervals, the attention weight tilts towards the time feature; when the biodiversity index is in different intervals, the attention weight is evenly distributed between the space-time features.

7. The method of claim 6, wherein the method further comprises: The clustering coefficient, in particular, refers to the density of mutual connection between node neighbors in the network, reflecting the local connectivity characteristics of the ecological network. The ratio of the actual number of connections between node neighbors to the total number of possible connections between neighbors is used to measure the local clustering characteristics of the network.

8. The method for assessing the effect of ecological restoration in three dimensions according to claim 7, characterized in that, The monitoring processing frequency, in particular, refers to the time interval period of data collection, sample analysis and index calculation in the ecological restoration area. By adjusting the monitoring frequency, the differential monitoring resource allocation of different ecological state regions is realized. When the salt marsh vegetation cover value is in different intervals, the corresponding monitoring frequency adjustment strategy is adopted.

Images