Land ecological protection and restoration effect dynamic evaluation system

Abstract

The invention relates to the technical field of land ecology assessment, and discloses a land ecology protection and restoration effect dynamic assessment system. According to the system, an ecological data acquisition unit is used for acquiring land utilization type data, vegetation coverage data, soil physicochemical property data and hydrogeological data of a target area and integrating the data into an ecological basic database; the dynamic feature extraction unit is used for carrying out space-time correlation analysis on various data in the ecological basic database and extracting a dynamic feature index set reflecting a land ecological state change trend; the restoration effect modeling unit is used for constructing a multi-scale evolution model of the land ecological restoration process according to the dynamic characteristic index set and generating a restoration effect evaluation parameter sequence; and the abnormal state recognition unit is used for establishing an ecological state deviation degree detection mechanism based on the restoration effect evaluation parameter sequence and outputting a spatial distribution map of the land ecological abnormal region. The system realizes continuous monitoring and quantitative evaluation of the ecological state of the land.

Description

Technical Field

[0001] This invention relates to the field of land ecological assessment technology, specifically a dynamic assessment system for the effectiveness of land ecological protection and restoration. Background Technology

[0002] Land is a fundamental resource for human survival and development, and its ecological condition directly affects regional ecological security, agricultural production safety, and the quality of the living environment. With economic and social development, accelerated industrialization and urbanization, and irrational land use, problems such as land degradation and impaired ecological functions are becoming increasingly prominent, including soil erosion, declining fertility, salinization, and biodiversity loss. To address these issues, land ecological protection and restoration projects are widely implemented globally. However, how to scientifically, objectively, and dynamically assess the actual effectiveness of these protection and restoration measures is a significant challenge facing current ecological management. Traditional land ecological assessment methods largely rely on periodic field surveys and sampling analyses, such as measuring vegetation cover and collecting soil samples for laboratory analysis. This method has significant limitations: long assessment cycles and high costs, making it difficult to achieve large-scale, high-frequency monitoring; data sources are scattered, with inconsistent collection times and spatial scales for different indicators, making effective correlation analysis difficult; and assessment results are mostly static snapshots, failing to reflect the dynamic changes and trends of the ecosystem.

[0003] While existing remote sensing technologies can monitor large-scale land cover changes, they often focus on analyzing single indicators and lack a comprehensive consideration of the land ecosystem. The land ecosystem is a complex system composed of multiple elements, including soil, vegetation, hydrology, and microorganisms, and its changes are the result of the interactions and influences of these elements. Improvement in a single indicator does not necessarily represent the restoration of the entire ecosystem's function. For example, increased vegetation cover may be due to short-term precipitation, but key indicators such as soil structure and organic matter content may not have improved simultaneously. Furthermore, ecological restoration is a long-term, dynamic process, and its effects are characterized by lag and fluctuation. Currently, there is a lack of dynamic assessment technologies that can integrate multi-source ecological data, reveal the system's inherent evolutionary patterns, and identify abnormal states in real time. Therefore, there is an urgent need for a land ecological protection and restoration effectiveness assessment system that can integrate multi-dimensional ecological indicators, characterize the system's dynamic evolution, and provide timely early warnings of ecological risks, providing accurate and timely scientific basis for ecological management decisions. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic evaluation system for the effectiveness of land ecological protection and restoration, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a dynamic evaluation system for the effectiveness of land ecological protection and restoration, the system comprising: The ecological data acquisition unit is used to acquire land use type data, vegetation cover data, soil physicochemical property data and hydrogeological data of the target area, and integrate them into an ecological basic database. The dynamic feature extraction unit is used to perform spatiotemporal correlation analysis on various types of data in the ecological basic database and extract a set of dynamic feature indicators that reflect the changing trends of land ecological status. The restoration effect modeling unit is used to construct a multi-scale evolution model of the land ecological restoration process based on a dynamic feature index set, and generate a sequence of restoration effect evaluation parameters. Anomaly identification unit is used to establish an ecological state deviation detection mechanism based on the restoration effect evaluation parameter sequence and output a spatial distribution map of land ecological anomaly areas.

[0006] Preferably, the ecological data acquisition unit acquires vegetation cover data in the following specific way: Vegetation indices were retrieved from multispectral remote sensing images, and a pixel-scale correction model was established by combining ground-measured quadrat data. The spatial continuity of the corrected vegetation index data was reconstructed using spatiotemporal kriging interpolation. Based on the reconstruction results, vegetation cover levels are divided and the area proportion of each level is calculated to form a time-series variation matrix of vegetation cover data.

[0007] Preferably, the dynamic feature extraction unit performs the following when conducting spatiotemporal correlation analysis: Spatial autocorrelation models were established for pH value, organic matter content and heavy metal concentration in soil physicochemical properties data; Detect the lag correlation coefficient between groundwater level fluctuations and surface runoff in hydrogeological data; Land use type data are converted into landscape pattern indices, and their coupling and coordination degree with vegetation cover data is calculated.

[0008] Preferably, the process of generating the dynamic feature index set includes: Soil parameters with Moran index greater than the threshold in the spatial autocorrelation model were selected as key soil features; Screening hydrological interaction features with lag correlation coefficients exceeding the significance level; Extract landscape-vegetation synergistic variation features with continuously decreasing coupling coordination. Key soil characteristics, hydrological interaction characteristics, and landscape-vegetation synergistic variation characteristics are spliced ​​together into a dynamic feature index set according to time steps.

[0009] Preferably, when the repair effect modeling unit constructs a multi-scale evolution model: At the pixel scale, a random forest algorithm is used to predict vegetation restoration potential; Construct coupled differential equations for soil erosion and nutrient cycling at the watershed scale; Establish a game equilibrium model of ecological carrying capacity and restoration investment at the regional scale.

[0010] Preferably, the method for generating the repair effect evaluation parameter sequence is as follows: Spatial clustering is performed on the pixel-scale prediction results, and the recovery potential value of the cluster center points is extracted; The steady-state solution of the coupled differential equation is used as a parameter for the ecological balance of the watershed; Record the iterative trajectory of the Pareto optimal solution in the game equilibrium model; The three types of parameters are aligned and integrated according to the repair project stage.

[0011] Preferably, the process of establishing the ecological state deviation detection mechanism includes: Calculate the sliding window coefficient of variation for each indicator in the parameter sequence for evaluating the repair effect; Identify the set of spatial locations corresponding to periods of sudden increases in the coefficient of variation; Construct an adjacency matrix that reflects the propagation path of abnormal ecological parameters; The core anomaly propagation nodes are determined based on the eigenvectors of the adjacency matrix.

[0012] Preferably, the steps for generating the spatial distribution map are as follows: Use the core anomaly propagation node as a seed point for region growth. Anomaly levels were classified using a Dirichlet process hybrid model. A spatial topology map of the abnormal area is generated by overlaying historical restoration project boundary data; Mark the deviation type of the dominant ecological parameter corresponding to each abnormal area.

[0013] Preferably, the system further includes an ecological risk early warning unit, which operates as follows: The ecological risk index is calculated based on the expansion rate of anomalous regions in the spatial distribution map. When the ecological risk index exceeds a preset threshold, a tiered early warning signal is triggered. Automatically generate early warning reports that include the coordinates of priority areas and the degree of parameter anomalies.

[0014] Preferably, the triggering conditions for the graded early warning signal include: The ecological risk index of the core abnormal diffusion node corresponding to the Level 1 warning has increased for three consecutive periods. A Level II warning requires that the area of ​​abnormal regions exceeds a set proportion of the total area of ​​the watershed; The Level 3 warning is activated when the deviation of the dominant ecological parameters includes both soil heavy metal contamination and vegetation degradation.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention system achieves refined and dynamic assessment of land ecological protection and restoration effects by constructing a complete technology chain from data integration, feature extraction, model building to anomaly identification. The ecological data acquisition unit systematically collects multi-dimensional data such as land use, vegetation cover, soil properties, and hydrogeology, integrating them into a unified ecological foundation database. This breaks down data silos in traditional assessments, laying a solid data foundation for comprehensive evaluation. The dynamic feature extraction unit performs spatiotemporal correlation analysis on various data in the foundation database, focusing not only on the current status of individual indicators but also on uncovering the synergistic changes of various indicators in time and space. The extracted dynamic feature indicator set can more profoundly reveal the overall evolution trend and internal driving mechanisms of the land ecosystem.

[0016] The restoration effect modeling unit constructs a multi-scale evolutionary model based on a dynamic feature index set. This model can simulate and predict the response process and future trend of land ecological status under different protection and restoration measures. The generated restoration effect assessment parameter sequence enables quantitative description and process tracking of restoration effects, making the assessment conclusions more predictable and instructive. The anomaly identification unit establishes a deviation detection mechanism based on the assessment parameter sequence. It can keenly capture early signals of ecosystem states deviating from their normal trajectory and accurately locate the spatial location of anomalies, presenting them intuitively in the form of distribution maps. This early warning capability enables managers to promptly identify problem areas, analyze causes, and take targeted measures to prevent the escalation of ecological degradation. This system elevates land ecological assessment from static, single-indicator description to a dynamic, multi-indicator coupled system analysis level, shifting from lagging post-event assessment to in-process monitoring and early warning. This significantly improves the precision and proactive response capabilities of land ecological management, providing an effective technical tool for the sustainable use of land resources and the maintenance of ecological security. Attached Figure Description

[0017] Figure 1 A trend chart for the phased assessment of the overall effectiveness of ecological restoration; Figure 2 Workflow diagram for acquiring vegetation cover data for ecological data collection units; Figure 3 Workflow diagram for generating dynamic feature index sets; Figure 4 A comparison chart of parameter sensitivity indices for multi-scale evolution models. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a dynamic evaluation system for land ecological protection and restoration effects. The system includes an ecological data acquisition unit responsible for acquiring land use type data, vegetation cover data, soil physicochemical property data, and hydrogeological data for the target area. Land use type data is derived from high-resolution remote sensing image interpretation, combined with ground surveys to form classified patches. Vegetation cover data is obtained by inverting vegetation indices from multispectral remote sensing images and corrected using ground-measured quadrat data. Soil physicochemical property data includes parameters such as pH value, organic matter content, and heavy metal concentration, obtained through field sampling and laboratory analysis. Hydrogeological data covers information such as groundwater level and surface runoff, recorded through long-term observation at monitoring stations. All data are integrated into an ecological foundation database, stored and indexed using a spatiotemporal database management system to ensure data consistency and accessibility. A dynamic feature extraction unit performs spatiotemporal correlation analysis on various types of data in the ecological foundation database to extract a set of dynamic feature indicators reflecting the changing trends of land ecological status. Spatiotemporal correlation analysis involves the construction of spatial autocorrelation models, calculation of lag correlation coefficients, and transformation of landscape pattern indices. The dynamic feature index set, composed of key soil features, hydrological interaction features, and landscape-vegetation co-variation features pieced together over time, characterizes the evolution of ecological parameters. The restoration effect modeling unit constructs a multi-scale evolutionary model of the land ecological restoration process based on this dynamic feature index set. This multi-scale model covers pixel, watershed, and regional scales, employing machine learning algorithms, differential equations, and game theory methods for modeling, respectively. The restoration effect assessment parameter sequence is generated by fusing model outputs from different scales, providing quantitative indicators of restoration effectiveness. The anomaly identification unit establishes an ecological state deviation detection mechanism based on the restoration effect assessment parameter sequence. This mechanism identifies ecological parameter anomalies by calculating the sliding window coefficient of variation and constructing an adjacency matrix for anomaly propagation paths. The spatial distribution map outputs the location and level of land ecological anomaly areas, assisting decision-makers in locating problem areas.

[0020] Example 1: See Figure 2The ecological data acquisition unit obtains vegetation cover data and retrieves vegetation indices from multispectral remote sensing images. These images are sourced from satellite platforms such as the Landsat or Sentinel series. The image data includes blue, green, red, and near-infrared bands. The vegetation index selected is the Normalized Differential Vegetation Index (NDVI). NDVI is calculated by dividing the difference between the near-infrared and red bands by the sum of these two bands. The retrieval process requires radiometric calibration to convert the digital quantization values ​​from the satellite sensors into surface radiance values. Atmospheric correction uses the FLAASH model to eliminate the influence of atmospheric molecules, aerosols, and thin clouds. Ground-measured quadrat data were deployed in the target area using a systematic grid method, with each quadrat measuring 10 meters by 10 meters. Measurement parameters included vegetation canopy coverage percentage, leaf area index, and aboveground biomass dry weight. The pixel-scale calibration model was established using a multiple linear regression method, fitting the remotely sensed NDVI values ​​with the ground-measured vegetation coverage values. The independent variables of the regression model included NDVI values, topographic slope data, and elevation data. Model coefficients were estimated using the least squares method. Model validation employed the hold-out method, dividing the dataset into training and test sets. The training set was used to train the model, and the test set was used to calculate the coefficient of determination and root mean square error to evaluate model accuracy.

[0021] Spatiotemporal kriging interpolation is used to reconstruct the spatial continuity of corrected vegetation index data. As a geostatistical method, it considers the correlation of data in both spatial and temporal dimensions. An exponential variogram model is chosen, and nugget values, sill values, and range parameters are determined using maximum likelihood estimation. The interpolation process is based on sampled data, predicting vegetation cover values ​​for unsampled points to generate continuous surface vegetation cover raster data. The spatial resolution of the raster data is consistent with the original remote sensing image, and the temporal resolution is set to 16 days, corresponding to the Landsat satellite revisit cycle. Internationally accepted classification standards are used to classify vegetation cover levels, with low cover levels... The vegetation coverage level is defined as follows: less than 30% for medium coverage, between 30% and 60% for medium coverage, and greater than 60% for high coverage. The area percentage of each level is calculated using the regional statistical analysis function of raster data. The number of pixels at each time point for each vegetation coverage level is counted, and the percentage of pixels at each vegetation coverage level relative to the total number of pixels in the area is calculated, forming a time-series variation matrix of the vegetation coverage data. This time-series variation matrix is ​​a two-dimensional array where rows represent time points and columns represent the area percentage of different vegetation coverage levels. Data is stored in NetCDF format, which supports efficient storage and retrieval of multidimensional arrays.

[0022] When performing spatiotemporal correlation analysis, the dynamic feature extraction unit establishes spatial autocorrelation models for pH, organic matter content, and heavy metal concentration in the soil physicochemical property data. The soil physicochemical property data are obtained through field sampling, and the sampling points are laid out using a regular grid method. The grid size is set according to the size of the study area and the degree of soil variability. Soil samples are collected at a depth of 0-20 cm at each sampling point. The soil pH is measured using the potentiometric method, the soil organic matter content is measured using the potassium dichromate oxidation method, and the soil heavy metal concentration, such as lead, cadmium, and chromium, is measured using inductively coupled plasma mass spectrometry. The spatial autocorrelation model uses the global Moran index. The calculation of the global Moran index requires the construction of a spatial weight matrix, which is defined using the Queen adjacency rule, i.e., polygons sharing vertices or edges are considered adjacent. The significance test of the global Moran index uses the permutation test method, generating 999 random permutations to calculate pseudo-p values. The lag correlation coefficient between groundwater level fluctuations and surface runoff in hydrogeological data was detected. The hydrogeological data were obtained from long-term continuous observation records of monitoring stations. Groundwater level data was collected daily by pressure-type water level sensors installed in monitoring wells. Surface runoff data was measured by a combination of flow velocity meters and water level gauges installed on river cross-sections. Instantaneous flow was calculated using the velocity-area method. The lag correlation coefficient was calculated using time series cross-correlation analysis. Before analysis, the stationarity of the groundwater level time series and surface runoff time series was tested using the Augmented Dickey-Fuller test. Non-stationary time series were differentially processed to make them stationary. The correlation coefficients were calculated for different lag orders, ranging from 0 months to 12 months. The significance of the correlation coefficients was tested using the t-test. The t-statistic was calculated and the p-value was obtained by consulting the t-distribution table.

[0023] Land use type data was converted into landscape pattern indices. Land use type data was obtained by interpreting high-resolution remote sensing imagery. The interpretation process employed an object-oriented classification method, using a first-level classification system including six categories: cultivated land, forest land, grassland, water area, construction land, and unused land. Landscape pattern indices were calculated using Fragstats software, including patch density, edge density, and the Shannon diversity index. Patch density equals the number of patches divided by the total landscape area, and edge density equals the sum of patch perimeters divided by the total landscape area. The Shannon diversity index was calculated based on the area ratio of patch types. The coupling and coordination degree between landscape pattern indices and vegetation cover data was calculated. The coupling and coordination degree model included a coupling degree model and a coordination degree model. The coupling degree model calculated the interaction strength between vegetation cover and each landscape pattern index, while the coordination degree model calculated the level of coordinated development between vegetation cover and landscape pattern indices. The coupling and coordination degree is the geometric mean of the coupling degree and the coordination degree, with a value ranging from 0 to 1. A value closer to 1 indicates better coordination between vegetation cover and landscape pattern. The inversion of vegetation indices from multispectral remote sensing images in the ecological data acquisition unit involves image preprocessing steps, including radiometric calibration and atmospheric correction. Radiometric calibration uses gain and offset parameters provided in the satellite image metadata to convert digital quantization values ​​into radiance values. Atmospheric correction uses the FLAASH model, with input parameters including image acquisition time, center point latitude and longitude, average elevation, and atmospheric model type. Ground-measured quadrat data layout considers topographic factors and vegetation type distribution to ensure that the quadrat can represent the main topography and vegetation types in the region. Quadrat locations are recorded with latitude and longitude coordinates using a high-precision GPS receiver, and the measurement time is strictly synchronized with the satellite transit time, controlled within two days before and after. The pixel-scale correction model training dataset contains three consecutive years of remote sensing image data and corresponding ground measurement data. In addition to NDVI values, slope, and elevation, the model input features also include aspect and precipitation data. The model output is the corrected vegetation cover estimate, which is a continuous value between 0 and 1. The correction model is evaluated using an independent validation dataset. The calculated coefficient of determination is used to measure the model's ability to explain variation, and the root mean square error is used to measure the magnitude of the prediction error.

[0024] The variogram fitting process of the spatiotemporal kriging interpolation method uses the least squares method to optimize the parameters of the variogram model. The spatial range is determined according to the distribution range of the sampling points, the time step is set to months, and the interpolation results are output in spatiotemporal cube data format. Each grid cell contains a series of vegetation cover values ​​at a series of time points. After the vegetation cover level is divided, the area proportion is calculated using the raster calculator function of the geographic information system software. The area of ​​each level is counted through conditional query statements. The time series change matrix update mechanism is set to automatically run the calculation process once for each new period of data, and the updated matrix overwrites the old version of the data. The construction of the spatial autocorrelation model for the dynamic feature extraction unit requires row standardization of the spatial weight matrix so that the sum of the weights in each row is 1. The global Moran index is calculated based on the ratio of the cross-product term to the variance. The local Moran index is calculated to generate a LISA clustering map. The LISA clustering map shows the spatial clustering types with statistical significance, including high-high clustering, low-low clustering, high-low anomaly, and low-high anomaly. The lag correlation coefficient analysis of hydrogeological data involves the stationarity test of the time series. Non-stationary series are subjected to first-order or second-order differencing to make them stationary. The lag order is selected based on the Akaike information criterion, choosing the lag order that minimizes the AIC value. When converting the landscape pattern index, the land use raster data is resampled to a uniform resolution, such as 30 meters by 30 meters. The coupling coordination degree calculation requires the vegetation cover data and landscape pattern index data to be standardized first to eliminate the influence of dimensions. The standardization method adopts min-max standardization to transform the data to between 0 and 1.

[0025] In the data integration phase of the ecological data acquisition unit, the ecological foundation database utilizes the PostgreSQL relational database management system combined with PostGIS spatial extensions. The database tables include timestamp fields, spatial geometry fields, and data value fields. Data quality control includes outlier detection and missing value imputation. Outlier detection employs the three-standard-deviation method; values ​​exceeding the mean plus or minus three standard deviations are considered outliers. Missing value imputation uses linear interpolation or K-nearest neighbor interpolation. The temporal variation matrix of vegetation cover data supports time series analysis functions; for example, trend analysis uses the Mann-Kendall trend... Potential testing and periodic analysis employed Fourier transform. The spatiotemporal correlation analysis of the dynamic feature extraction unit output a set of dynamic feature indicators, which were stored as a three-dimensional array. The three dimensions represented time, spatial location, and feature indicators, respectively. The spatial autocorrelation model results of soil physicochemical properties were visualized using a heatmap, with color intensity representing the magnitude of the Moran index. The lag correlation coefficient matrix of hydrological interaction features was used to construct a directed network graph, where nodes represented hydrological elements and edges represented lag correlation relationships. The time series of landscape-vegetation synergistic variation features was used to detect ecological degradation trends, and linear regression analysis was used to calculate the slope of change.

[0026] Example 2: See Figure 3 The generation process of the dynamic feature index set includes selecting soil parameters with Moran's index greater than a threshold in the spatial autocorrelation model as key soil features. The spatial autocorrelation model is constructed based on soil physicochemical property data, which are derived from laboratory analysis results of field sampling points. The threshold is set using a statistical significance level, specifically a p-value less than 0.05. The significance test of Moran's index is achieved by calculating the standardized statistic Z-value; a Z-value greater than 1.96 is considered statistically significant. Only soil parameters with a Moran's index significantly greater than 0 are selected as key soil features. These features reflect the spatial clustering distribution pattern of soil properties. Hydrological interaction features with lag correlation coefficients exceeding the significance level are screened. The lag correlation coefficients are calculated by analyzing groundwater level fluctuation sequences and surface runoff sequences in hydrogeological data. The significance level is set at a p-value less than 0.01, and a two-tailed t-test is used to determine the significance of the correlation coefficients. Both positive and negative lag correlation coefficients exceeding the significance level are retained. These coefficients reveal the time-delay response relationship between groundwater level changes and surface runoff. We extracted the landscape-vegetation synergistic variation characteristics of continuously decreasing coupling coordination degree. The coupling coordination degree was calculated by analyzing the landscape pattern index converted from land use type data and vegetation cover data. The continuous decrease was defined as the coupling coordination degree value showing a monotonically decreasing trend over three consecutive observation periods. This decreasing trend indicates that the coordination relationship between landscape pattern and vegetation cover is deteriorating.

[0027] Key soil characteristics, hydrological interaction characteristics, and landscape-vegetation synergistic variation characteristics are concatenated into a dynamic feature index set according to a time step. The time step is consistent with the data update cycle of the ecological data acquisition unit, specifically set to a monthly cycle. The concatenation operation is aligned in the time dimension, with each time step corresponding to a feature vector containing the values ​​of all selected features. The final organization of the dynamic feature index set is a three-dimensional data structure, with the three dimensions corresponding to time, spatial location, and feature index number, respectively. Data storage adopts HDF5 format to support efficient read and write operations. The update mechanism of the dynamic feature index set is synchronized with the data acquisition of the ecological data acquisition unit. Each time new monitoring data is entered into the database, the features are recalculated. The feature calculation process adopts a distributed computing framework to improve processing efficiency. When constructing a multi-scale evolutionary model for vegetation restoration effect modeling, the random forest algorithm is used at the pixel scale to predict vegetation restoration potential. The random forest algorithm is an ensemble learning method that improves prediction accuracy by constructing multiple decision trees and voting on them. The feature variables used to train the random forest model include historical vegetation cover data, topographic factor data, climate factor data, and soil background data. Topographic factor data includes slope, aspect, and elevation; climate factor data includes monthly average temperature and total monthly precipitation; and soil background data includes pH and organic matter content. The model response variable is the future change in vegetation cover. After model training, the model is applied to each pixel, outputting the vegetation restoration potential value for that pixel at a specific future period. The restoration potential value is a probability value between 0 and 1, representing the likelihood of vegetation cover reaching an ideal state. The hyperparameters of the random forest model are determined using a grid search method, which traverses a predefined parameter space and selects the parameter combination that performs best on the validation set.

[0028] Coupled differential equations for soil erosion and nutrient cycling were constructed at the watershed scale. Soil erosion was described using the modified generalized soil loss equation (RUSLE), which includes rainfall erosivity, soil erodibility, topography, vegetation cover, and soil and water conservation measures. Nutrient cycling primarily considered the migration and transformation of nitrogen and phosphorus, establishing a nitrogen and phosphorus mass conservation equation. This equation included process terms related to plant uptake, soil fixation, leaching loss, and mineralization release. The soil erosion and nutrient cycling equations were coupled through changes in soil thickness and nutrient content. Parameters in the coupled differential equations were obtained through literature review and field experiments. Numerical methods, such as the fourth-order Runge-Kutta method, were used to solve the equations, with a calculation step size of one month. Boundary conditions for the coupled differential equations were set based on the actual geographical boundaries of the watershed, and initial conditions were initialized using field monitoring data.

[0029] A game equilibrium model of ecological carrying capacity and restoration input is established at the regional scale. Ecological carrying capacity is derived by comprehensively calculating the regional water resource carrying capacity, land resource carrying capacity, and environmental capacity carrying capacity. Restoration input is quantitatively represented as the weighted sum of capital input, labor input, and material input. Game participants include government departments, local communities, and environmental protection organizations. The objective function of government departments is to maximize overall benefits, the objective function of local communities is to maximize livelihood improvement, and the objective function of environmental protection organizations is to maximize ecological value. The game equilibrium model adopts a non-cooperative game framework to solve for the Nash equilibrium point. The Nash equilibrium point represents the state where, given the strategies of other participants, no participant can obtain a higher payoff by unilaterally changing their strategy. The payoff matrix of the game equilibrium model is determined through expert scoring and historical data regression analysis. The integration of multi-scale evolutionary models is achieved by defining scale transformation rules. The predicted vegetation restoration potential at the pixel scale is upscaled to the watershed scale using spatial aggregation methods, such as arithmetic mean or area-weighted average. The output of the coupled differential equations of soil erosion and nutrient cycling at the watershed scale serves as a constraint on the regional-scale game equilibrium model. For example, if the watershed soil erosion modulus exceeds the allowable value, it will reduce the regional ecological carrying capacity. The output of the game equilibrium model is fed back to the pixel scale, influencing the priority setting of restoration measures at the pixel scale. This cross-scale feedback mechanism is achieved through iterative calculations, iterating until the model results converge. The scale transformation rules consider the spatial heterogeneity between different scales, employing a combination of upscaling and downscaling methods. The dynamic feature index set serves as the input data for the multi-scale evolution model. Key soil features are used to initialize soil erodibility factors and nutrient background values. Hydrological interaction features are used to calibrate watershed hydrological model parameters. Landscape-vegetation covariance features are used to assess the spatial heterogeneity of vegetation restoration potential. The model's operating cycle corresponds to the stages of ecological restoration projects, including the planning, implementation, and consolidation phases, with corresponding evaluation parameters output for each stage. Preprocessing of the model input data includes outlier detection and missing value imputation. Outlier detection employs box plots, and missing value imputation uses multiple interpolation.

[0030] The method for generating the parameter sequence for evaluating the restoration effect involves spatial clustering of the pixel-scale prediction results. The K-means algorithm is selected for spatial clustering, and the number of clusters is determined according to the elbow rule. The elbow rule calculates the sum of squares within each cluster corresponding to different numbers of clusters, selects the number of clusters corresponding to inflection points, extracts the coordinates of the center points of each cluster, and reads the vegetation restoration potential values ​​of these center points in the random forest model prediction results. These values ​​represent the typical restoration potential of different landscape types. The steady-state solution of the coupled differential equations is used as the watershed ecological balance parameter. The steady-state solution is obtained by setting the time derivative to zero and solving the algebraic equation system. The Newton-Raphson iteration method is used to solve the algebraic equation system, and the iteration convergence condition is set to the relative error between two adjacent iterations being less than 0.001. The iteration trajectory of the Pareto optimal solution in the game equilibrium model is recorded. The Pareto optimal solution is the set of solutions that cannot further improve the interests of one party without harming the interests of any party. The iteration trajectory records the changes of each objective function value in each iteration, and the trajectory data is used to analyze the convergence of the game process. The three types of parameters were aligned and fused according to the repair project stage. The alignment operation was based on timestamp matching. The fusion method adopted was principal component analysis for dimensionality reduction followed by weighted summation. The weights were allocated according to the importance of each parameter in the overall evaluation, and the weight values ​​were determined by the analytic hierarchy process.

[0031] The updating of the dynamic feature index set triggers the recalibration of the multi-scale evolutionary model. When new monitoring data indicates significant changes in feature values, the model parameters need to be re-estimated. The criterion for significant changes is that the change in feature values ​​relative to the baseline value exceeds 10%. The model calibration uses the maximum likelihood estimation method. The update of the pixel-scale random forest model is achieved through an online learning mechanism. When new vegetation cover observation data is obtained, the model uses the new data to fine-tune the existing decision tree. The fine-tuning process restricts the structure of the decision tree from undergoing significant changes, and the gradient boosting method is used to gradually adjust the model parameters. Sensitivity analysis of parameters in the watershed-scale differential equation helps identify key parameters. The sensitivity analysis uses the Morris screening method, and parameters with high sensitivity are calibrated more frequently, with the calibration frequency set to once a month. The utility function parameters of the regional-scale game equilibrium model are determined through an expert questionnaire survey. The questionnaire survey uses the Delphi method, and multiple rounds of feedback are conducted until the expert opinions converge. The convergence criterion for the expert opinions is a coefficient of variation of less than 0.1.

[0032] Uncertainty propagation analysis of the multi-scale evolution model employs Monte Carlo simulation, specifying a probability distribution (e.g., normal or uniform) for each input parameter. The distribution range of the output results is simulated through multiple random samplings. The uncertainty analysis results are presented in the form of confidence intervals, with a confidence level set to 95%. Model validation uses independent datasets, comparing model predictions with actual observational data not involved in the modeling. Mean squared error and Nash coefficient are calculated to evaluate model performance; a Nash coefficient greater than 0.7 indicates acceptable model performance. Visualization of model results uses multi-dimensional charts combined with Geographic Information System (GIS) maps to dynamically display the spatiotemporal evolution of restoration effects at different scales. The visualization system supports multi-dimensional data filtering and dynamic playback. Storage and management of the restoration effect evaluation parameter sequences utilize a time-series database structure. The database index is built based on timestamps and spatial coordinates, supporting fast querying and retrieval. Anomaly detection in the parameter sequences employs a change point analysis algorithm. This algorithm identifies structural change points in the sequence based on Bayesian information principles, with these change points corresponding to important turning points in the ecological restoration project. The long-term trend analysis of the parameter series adopts the seasonal decomposition method, which decomposes the series into trend components, seasonal components and residual components. The trend components are used to assess the long-term direction of change of the repair effect.

[0033] See Figure 4 A comparison chart of parameter sensitivity indices for multi-scale evolutionary models is presented. In the parameter sensitivity analysis of multi-scale evolutionary models, the Morris screening method was used to quantify the influence of each parameter on the model output. Specifically, the sensitivity indices of six key parameters—soil erodibility, rainfall erosivity, vegetation cover, topography, nitrogen cycle coefficient, and phosphorus cycle coefficient—were evaluated at both the watershed and pixel scales. The sensitivity indices were calculated by repeatedly perturbing the parameter values ​​and observing the coefficient of variation of the model output. The indices range from 0.0 to 1.0, with higher values ​​indicating stronger sensitivity to the model results. In the chart, the length of the bars visually reflects the magnitude of parameter sensitivity; dark gray represents the watershed scale, and light gray represents the pixel scale. Regarding parameter configuration, the sensitivity analysis was based on model operation results, with the perturbation amplitude set at 10% of the parameter standard deviation and 1000 sampling times to ensure statistical robustness. The analysis shows that soil erodibility and rainfall erosivity are highly sensitive at the watershed scale (index 0.7), while vegetation cover is relatively sensitive at the pixel scale (index 0.6), revealing differences in dominant ecological processes at different scales. Sensitivity results are used to optimize model calibration frequency; high-sensitivity parameters are recalibrated monthly, while low-sensitivity parameters are adjusted quarterly.

[0034] Example 3: The method for generating the parameter sequence for evaluating the restoration effect is to perform spatial clustering on the pixel-scale prediction results. The pixel-scale prediction results are derived from the estimation of vegetation restoration potential by the random forest algorithm. Each pixel corresponds to a 30m × 30m surface area. The random forest algorithm uses 200 decision trees with a maximum tree depth of 20 layers. The spatial clustering algorithm selected is the K-means clustering algorithm. The number of clusters is determined by the elbow rule. The elbow rule calculates the sum of squares within each cluster when the number of clusters ranges from 2 to 15. The formula for calculating the sum of squares within a cluster is the sum of the squares of the Euclidean distances from all pixels to the center of their respective clusters. The number of clusters corresponding to the inflection point where the rate of decrease in the sum of squares within a cluster begins to slow down is selected as the optimal number of clusters. The restoration potential value of cluster centers is extracted. The cluster centers are obtained by calculating the arithmetic mean of the coordinates of all cells in each spatial cluster. The vegetation restoration potential value of the center point is the weighted average of the restoration potential values ​​of all cells in the spatial cluster. The weights are determined according to the cell area ratio. These center point restoration potential values ​​constitute a subset of evaluation parameters at the cell scale. The parameter subset is organized and stored according to the cluster number, and the storage format is geospatial data exchange format.

[0035] The steady-state solution of the coupled differential equations is used as parameters for the watershed's ecological balance. The coupled differential equations describe the interaction between soil erosion and nutrient cycling. The soil erosion process uses the modified general soil loss equation, and the nutrient cycling process uses the nitrogen and phosphorus mass conservation equation. The steady-state solution transforms the differential equations into a system of nonlinear algebraic equations by setting the time derivatives of the soil thickness change equation and the nutrient mass conservation equation to zero. The solution of the nonlinear algebraic equations uses the Newton-Raphson iteration method. The initial values ​​of the iterations are the average values ​​of measured data from monitoring stations within the watershed. The iteration step size adopts an adaptive adjustment strategy, and the convergence condition is set to the relative error between two adjacent iterations being less than one-thousandth. The steady-state solution includes two parameters: soil equilibrium thickness and nutrient equilibrium concentration, with units of millimeters and milligrams per kilogram, respectively. This study records the iterative trajectory of the Pareto optimal solution in a game equilibrium model involving three participants: a government department, a local community, and an environmental protection organization. Each participant's utility function is in the form of a Cobb-Douglas function. The Pareto optimal solution is obtained using a non-dominated sorting genetic algorithm with a population size of 100 individuals, 200 generations, a crossover probability of 0.9, and a mutation probability of 0.1. The iterative trajectory records the objective function values ​​of the Pareto solution set in each generation. The objective function includes ecological benefit indices, economic benefit indices, and social benefit indices. The trajectory data is stored in time-series format, with a sampling frequency consistent with the number of iterations.

[0036] The three types of parameters were aligned and fused according to the repair project stages, which were divided into three phases: planning, implementation, and consolidation. The duration of each phase was determined based on the specific project. The alignment operation was based on a unified timestamp matching, with a time window size of three months. The fusion method employed principal component analysis (PCA) for dimensionality reduction followed by weighted summation. PCA extracted principal components with a cumulative contribution rate exceeding 85%, and the weights were determined using the analytic hierarchy process (AHP). The AHP judgment matrix was constructed using the nine-scale method.

[0037] The weighted summation formula is expressed as:

[0038] in: This represents the integrated evaluation parameters after fusion. The weighting coefficients represent the pixel scale parameters. This represents the pixel scale parameter value. This represents the weighting coefficients of the watershed-scale parameters. Indicates watershed-scale parameter values. This represents the weighting coefficients of the regional scale parameter. This represents the regional scale parameter value. The weighting coefficients satisfy the normalization condition. The weight values ​​are obtained by calculating the eigenvector corresponding to the largest eigenvalue of the judgment matrix.

[0039] The parameter sequences for evaluating restoration effectiveness are stored using a time-series database structure. The database index is built based on timestamps and spatial encoding, supporting fast queries by time and spatial range. Anomaly detection of the parameter sequences employs a change point analysis algorithm. This algorithm identifies structural change points in the sequence based on Bayesian information criteria, with these change points corresponding to important time nodes in the ecological restoration project. Long-term trend analysis of the parameter sequences uses a seasonal decomposition method, decomposing the sequence into trend components, seasonal components, and residual components. The trend component is used to assess the sustainability of the restoration effect. Spatial clustering of pixel-scale prediction results considers geographical constraints. The spatial clustering algorithm introduces a spatial adjacency matrix as a constraint term to ensure spatial continuity of pixels within each spatial cluster. Visualization of the clustering results uses a color coding method, assigning different colors to different spatial clusters, with cluster boundaries superimposed on the digital elevation model. The solution process for the watershed-scale steady-state solution includes parameter sensitivity analysis. This sensitivity analysis uses a local differential method to calculate the degree of influence of each parameter on the steady-state solution. The analysis of the iterative trajectory of the game equilibrium model includes a convergence test, which is determined by calculating the Hausdorff distance between solution sets in successive generations.

[0040] The time window size for parameter alignment and fusion is determined through cross-validation, which divides the data into training and test sets and selects the time window size that minimizes the prediction error. The dimensionality reduction of principal component analysis (PCA) is determined using scree plots, which display the variance contribution rate of each principal component; the number of dimensions before a sharp decrease in variance contribution rate is selected. The determination of weight coefficients includes a consistency check; a consistency ratio less than 0.1 indicates that the judgment matrix meets the consistency requirements. The update mechanism for the parameter sequence for evaluating the repair effect is synchronized with the repair project progress, triggering parameter recalculation after each project phase. Quality control of the parameter sequence includes outlier detection, which uses the isolated forest algorithm to identify parameter values ​​significantly different from other data points. Standardization of the parameter sequence uses the Z-score method to transform parameters of different dimensions to the same scale. Missing values ​​in the parameter sequence are handled using multiple imputation, generating multiple complete datasets for statistical analysis. The spatial representativeness of the pixel-scale parameter subset is verified by calculating the spatial autocorrelation index, using the Moran index; a Moran index greater than 0 indicates spatial clustering of the parameters. The physical rationality of the watershed-scale parameters was verified through extreme value testing, and the parameter values ​​should be within the theoretically reasonable range. The social feasibility of the regional-scale parameters was verified through expert review, which employed the Delphi method for multiple rounds of back-to-back evaluation. The robustness of the parameter fusion results was verified through Monte Carlo simulation, simulating the impact of random perturbations in the input parameters on the fusion results.

[0041] The application of the remediation effectiveness evaluation parameter sequence includes trend prediction functionality, employing an autoregressive integral moving average model with model parameters determined through maximum likelihood estimation. Comparative analysis of the parameter sequences includes horizontal and vertical comparisons: horizontal comparisons of parameter differences across different remediation areas, and vertical comparisons of parameter changes within the same area over different periods. Parameter sequence report generation utilizes a template-based approach, with templates including parameter tables, trend charts, and spatial distribution maps. The parameter sequence access interface includes an application programming interface (API), supporting third-party system calls and data exchange. The generation process of the remediation effectiveness evaluation parameter sequence is automated, requiring no manual intervention from data input to result output. The pipeline comprises a data validation module, a computation engine module, and a result output module, communicating via a message queue. The computation engine module employs a distributed computing framework, supporting large-scale parallel data processing. The result output module supports multiple data formats, including JSON, CSV, and NetCDF. The entire pipeline's operational status is tracked in real-time by a monitoring system, with automatic alarm mechanisms triggered for abnormal situations.

[0042] Example 4: The establishment process of the ecological state deviation detection mechanism includes calculating the sliding window coefficient of variation (COP) of each indicator in the restoration effect assessment parameter sequence. The restoration effect assessment parameter sequence is derived from the output of a multi-scale evolution model, including pixel-scale vegetation restoration potential, watershed-scale ecological balance parameters, and regional-scale game equilibrium parameters. The sliding window duration is set to 12 months, and the window sliding step size is set to 1 month. The COP is calculated using the ratio of standard deviation to mean. The COP is calculated independently for each parameter within each window, generating a COP time series. The COP time series is stored using a time series database structure, supporting fast retrieval by time range and parameter type. The spatial location set corresponding to periods of sudden COP increases is identified. The criterion for a sudden COP period is that the COP value exceeds twice the standard deviation of the historical mean. The historical mean is calculated based on data from the previous 36 months. The spatial location set maps the location identifiers in the parameter sequence to actual geographic coordinates through geocoding. The coordinate system used is the National Geodetic Coordinate System. A spatial point set is generated for each sudden COP period, containing all geographic locations where COP sudden increases occurred within that period. Spatial analysis of the spatial location set employs a buffer analysis method, with a buffer radius set to 1000 meters, to determine the scope of anomaly impact.

[0043] An adjacency matrix reflecting the propagation path of ecological parameter anomalies is constructed. The rows and columns of the adjacency matrix correspond to all possible spatially located nodes, and the element values ​​represent the propagation intensity of ecological parameter anomalies between these nodes. Propagation intensity is calculated based on the spatial distance decay law and the correlation of ecological processes, with the distance decay coefficient set to 0.01. The correlation of ecological processes is quantified using canonical correlation analysis. The construction of the adjacency matrix considers multiple propagation pathways, including hydrological connectivity, species migration paths, and anthropogenic disturbance channels. The adjacency matrix is ​​stored in a sparse matrix format to reduce storage space requirements. Core anomaly propagation nodes are identified based on the eigenvectors of the adjacency matrix. These eigenvectors are obtained by calculating the principal eigenvectors of the adjacency matrix using a power iteration method, with the principal eigenvector corresponding to the largest eigenvalue. The criterion for identifying core anomaly propagation nodes is that the element values ​​in the eigenvectors are greater than twice the standard deviation of the mean. These nodes occupy a central position in the ecological anomaly propagation network and have the strongest propagation influence. The spatial distribution pattern of core anomaly propagation nodes is visualized using kernel density analysis.

[0044] The spatial distribution map generation process involves using core anomaly diffusion nodes as seed points for region growth. A grid-based flood-fill algorithm is employed for this region growth, with growth criteria based on spatial adjacency and ecological parameter similarity exceeding a threshold of 0.8. Euclidean distance is used for similarity calculation, and the growth direction considers eight neighborhood directions. Growth terminates when an ecological type boundary is encountered or the maximum growth radius of 5000 meters is reached. The geometric accuracy of the region growth results is evaluated using root mean square error to ensure accurate boundary positioning. A Dirichlet process mixture model is used to classify anomaly levels. The concentration parameter of the Dirichlet process mixture model is set to 1.0, the base distribution is a Gaussian distribution, and the number of iterations is set to 1000. Anomalies are classified into three levels based on posterior probability distribution, with each anomaly region assigned to the level with the highest probability. Spatial consistency is checked to eliminate isolated anomaly classification units. Historical restoration project boundary data, sourced from an engineering project archive, is overlaid to generate a spatial topological map of the anomaly regions. This data includes polygonal boundaries and engineering attribute information. Spatial topological relationship calculation includes determining relationships such as inclusion, adjacency, and disjointness. The topological relationship graph is stored using a graph structure, where nodes represent abnormal regions and edges represent topological relationships. The layout algorithm for the topological relationship graph uses a force-directed graph algorithm to optimize visualization.

[0045] The deviation types of the dominant ecological parameters corresponding to each anomaly area are labeled. The dominant ecological parameters are determined by comparing the degree of deviation of each ecological parameter, and the degree of deviation is calculated using standardized anomaly values. The labeled information is written into a spatial attribute table, which, together with the geometric data of the anomaly areas, constitutes a complete spatial distribution map. The coordinate reference system of the spatial distribution map is uniformly converted to a geographic coordinate system to facilitate cross-platform data sharing.

[0046] Table 1: Criteria for Classifying Anomalies

[0047] The implementation of the ecological state deviation detection mechanism requires a dedicated computing environment, including a 64-core CPU processor, 256GB of RAM, and 2TB of solid-state drive storage. The software environment uses a Python data science stack, primarily relying on NumPy, SciPy, and Scikit-learn libraries. The detection mechanism is set to run automatically once a month, during off-peak hours when system load is low. Monitoring of the computing environment involves real-time tracking of resource usage using system performance counters.

[0048] The coefficient of variation (COP) is calculated using a sliding window algorithm, implemented using the `rolling` function from the Pandas library, with a rectangular window selected as the window type. Sudden increases in COP are detected using a mutation detection algorithm employing the CUSUM cumulative sum control chart method, with control limits set to three standard deviations. Spatial location sets are stored using the point data model of the PostGIS spatial database, supporting spatial queries and analysis. An R-tree index is used to index the spatial location sets, accelerating spatial queries. The adjacency matrix construction process includes matrix sparsification, retaining the top 10% of edges by weight to reduce storage space and improve computational efficiency. Identification of core anomaly propagation nodes utilizes network centrality analysis, including degree centrality, proximity centrality, and eigenvector centrality calculations. Node importance ranking employs the PageRank algorithm, with a damping coefficient of 0.85 and an iterative convergence tolerance of 0.0001. Network analysis results are visualized and validated using Gephi software.

[0049] The implementation of the region growing algorithm includes growth rate control, which expands outward by only one pixel in each iteration. The inference of the Dirichlet process mixture model employs the Gibbs sampling method, with the Gibbs sampling burn-in period set to the first 200 iterations and the sampling interval set to 5 iterations. The anomaly level classification results undergo spatial smoothing using a majority filtering method with a 3×3 pixel filter window. The number of iterations for spatial smoothing is set to 3 to ensure result stability. The generation of the spatial topology map utilizes spatial predicate computation from a spatial database, including topological relationship judgments such as Within, Contains, and Intersects. Principal component analysis (PCA) is used to determine the deviation types of dominant ecological parameters, extracting the loading coefficient of the first principal component as the parameter importance weight. The visualization of the spatial distribution map uses a hierarchical color scheme, with different anomaly levels represented by different color intensities, and dominant deviation types distinguished by pattern filling. The visualization legend configuration follows international cartographic standards to ensure clear color semantics.

[0050] The quality control of the ecological state deviation detection mechanism includes accuracy verification, which employs cross-validation by randomly dividing the data into training and test sets to calculate the detection accuracy. The mechanism's operational monitoring includes performance indicator recording, encompassing computation time, memory usage, and result consistency. Updates and maintenance include parameter optimization, using a grid search method to find the optimal parameter combination. The mechanism's output includes detailed log records, recording runtime, input parameters, intermediate results, and final conclusions. The spatial distribution map application includes a dynamic update mechanism that triggers local recalculation when new monitoring data is added, improving update efficiency. The spatial distribution map publishing service uses the OGC standard map service, supporting both WMS and WFS interface protocols. The spatial distribution map analysis functions include spatial statistics, such as hotspot analysis, spatial autocorrelation analysis, and spatial interpolation analysis. The interactive query function supports multi-condition queries, including time range, spatial range, and anomaly level.

[0051] Example 5: The ecological risk early warning unit calculates the ecological risk index based on the expansion rate of anomalous areas in the spatial distribution map. The spatial distribution map is derived from the analysis results of the ecological state deviation detection mechanism and includes the spatial boundaries and attribute information of anomalous areas at different levels. The expansion rate is calculated by comparing the area changes of anomalous areas in adjacent monitoring periods. Specifically, a raster data overlay analysis method is used to compare the raster images of anomalous areas in two periods at the pixel level and count the ratio of newly added anomalous pixels to the original number of anomalous pixels. The monitoring period is set to a quarterly cycle. When calculating the area change, the weight difference of different anomalous levels is considered: the weight of level 1 anomalous areas is set to 1.0, the weight of level 2 anomalous areas is set to 0.7, and the weight of level 3 anomalous areas is set to 0.4. The ecological risk index is calculated using a weighted summation model. The model input includes three indicators: the absolute value of the expansion rate, the relative rate of change, and spatial connectivity, with weight coefficients set to 0.4, 0.35, and 0.25, respectively. When the ecological risk index exceeds a preset threshold, a tiered early warning signal is triggered. The preset thresholds are set based on historical data distribution, using the 85th percentile of the ecological risk index data from the past five years as the Level 1 warning threshold, the 70th percentile as the Level 2 warning threshold, and the 50th percentile as the Level 3 warning threshold. The tiered early warning signals are divided into three levels: red, orange, and yellow. Red corresponds to Level 1, orange to Level 2, and yellow to Level 3. Each warning level corresponds to a different emergency response procedure: a red warning requires activation within 24 hours, an orange warning within 48 hours, and a yellow warning within 72 hours. After the warning signal is triggered, it is simultaneously disseminated through multiple methods: the monitoring center control room activates the audible and visual alarm device, the system automatically sends warning SMS messages to the mobile phones of relevant responsible persons, and the government affairs office platform updates the warning status simultaneously.

[0052] The system automatically generates early warning reports containing the coordinates of priority response areas and the degree of parameter anomalies. Priority response areas are determined based on the ecological risk index, with the top five areas having the highest index selected. Parameter anomalies are quantified using standardized anomalies, calculating the deviation multiple between the current parameter value and the historical average for the same period. The early warning report adopts a standardized document structure, including five parts: early warning overview, data analysis, spatial distribution, risk assessment, and response recommendations. After generation, the report is automatically distributed to water resources departments, ecological environment departments, and local government emergency management departments through the government information system. Important early warnings are also copied to higher-level authorities. The triggering conditions for tiered early warning signals include three independent criteria. The first criterion is a continuous increase in the ecological risk index of the core anomaly diffusion node for three consecutive periods. The core anomaly diffusion node is identified by the ecological state deviation detection mechanism, and "three consecutive periods" refers to three consecutive quarterly monitoring cycles. The second criterion is that the area of ​​the anomaly zone exceeds a set percentage of the total watershed area. This set percentage is dynamically adjusted based on the watershed's ecological sensitivity, set at 5% for ecologically sensitive watersheds and 10% for general watersheds. The third criterion is that the deviation type of the dominant ecological parameter includes both soil heavy metal exceedance and vegetation degradation. Soil heavy metal exceedance is judged according to the "Soil Environmental Quality Agricultural Land Soil Pollution Risk Control Standard", and vegetation degradation is judged according to the quarterly decline rate of vegetation coverage exceeding 10%.

[0053] The implementation of the ecological risk early warning unit is illustrated using monitoring data from a tributary of the Yangtze River in 2023. The spatial distribution map shows one primary anomaly area in the northern phosphate mining area, and one primary anomaly area and four secondary anomaly areas in the southern hilly agricultural area, mainly distributed along the riverbanks and around the mining area. The expansion rate of the anomaly areas was calculated using a geographic information system overlay analysis tool, comparing the changes in anomaly areas between the first and second quarters of 2023. The calculation results show that the expansion rate of the anomaly areas around the northern mining area reached 15% of the quarterly growth rate, while the expansion rate of the anomaly areas in the southern agricultural area was 8%. In the ecological risk index calculation, the ecological risk index for the northern mining area was 0.78, and the ecological risk index for the southern agricultural area was 0.62. The warning thresholds were set based on historical monitoring data from 2018 to 2022 for a tributary region of the Yangtze River. Statistical analysis determined the red warning threshold to be 0.75 for the ecological risk index, the orange warning threshold to be 0.65, and the yellow warning threshold to be 0.55. Monitoring data from the second quarter of 2023 showed that the ecological risk index of the core abnormal diffusion node in the northern mining area rose from 0.68 to 0.79 for three consecutive quarters, triggering a red warning. Simultaneously, the abnormal area of ​​the core node accounted for 6% of the total basin area, triggering an orange warning. Specialized testing of the core node area confirmed that the soil cadmium content exceeded the standard by 2.1 times, and the quarterly vegetation coverage decline rate reached 13%, meeting the yellow warning trigger conditions. During the automatic generation of the warning report, the priority disposal area identification algorithm marked areas with an ecological risk index greater than 0.7 as red warning zones. Parameter anomaly analysis showed that the soil organic matter content deviated from the benchmark value by 38%, and the groundwater level decline deviated from the benchmark value by 25%. The warning report was automatically pushed to relevant departments through the basin ecological management platform. The report included a detailed spatial distribution map, data statistics tables, and a list of disposal recommendations.

[0054] The hardware configuration of the ecological risk early warning unit includes a multi-screen monitoring workstation, audible and visual alarm devices, and network communication equipment. The monitoring workstation is equipped with a high-performance graphics processor, supporting real-time rendering of large-scale spatial data. The audible and visual alarm devices use a combination of a red rotating light and a buzzer, and the network communication equipment supports dual-link backup via wired and wireless networks. The software system adopts a microservice architecture, independently deploying functional modules such as risk calculation, early warning determination, and report generation, and exchanging data through message queues. The early warning response mechanism establishes a multi-departmental collaborative workflow. A red alert triggers a bureau-level emergency response, establishing an on-site command center for unified dispatch. An orange alert triggers a division-level emergency response, designating a specific person to coordinate and handle the situation. A yellow alert triggers a section-level emergency response, requiring the responsible unit to rectify the situation within a specified timeframe. Emergency resource dispatch implements a hierarchical management system, configuring corresponding monitoring equipment, disposal materials, and professional teams according to the early warning level.

[0055] The ecological risk early warning unit achieves standardized data interface connectivity with external systems. Meteorological data is integrated into the China Meteorological Administration's real-time forecast system, with weather forecasts updated hourly. Hydrological data is connected to the Ministry of Water Resources' national hydrological monitoring platform, transmitting water level and flow monitoring data in real time. Pollution source data is integrated into the ecological and environmental protection department's key pollution source online monitoring system, acquiring enterprise emission data in real time. Secure data transmission employs SSL encryption, and a firewall isolation mechanism protects system security. Early warning information is disseminated through government intranets, mobile terminals, and public information platforms, ensuring information coverage of all relevant parties. Detailed early warning reports are published on the government intranet, simplified early warning information is pushed to mobile terminals, and basic early warning information is publicized on public information platforms.

[0056] The ecological risk early warning unit successfully issued an early warning for an ecological risk event that occurred in a mineral resource development zone in August 2023, issuing an orange alert 16 days in advance. The warning report indicated an increased risk of heavy metal diffusion in the soil surrounding the tailings dam and recommended immediate anti-seepage reinforcement measures. Relevant departments responded promptly based on the warning information, organizing professional teams for emergency response to prevent a major ecological accident. Post-event assessment showed that the early warning accuracy rate reached 89%, detecting risk signs 22 days earlier than traditional monitoring methods. Continuous improvement of the early warning system includes establishing a user feedback mechanism, collecting feedback from management departments quarterly. Model parameters are regularly optimized, and machine learning algorithms are used to automatically adjust weight coefficients. The system version is iteratively updated, with two major versions released annually to maintain technological advancement. Standard interfaces are established for data sharing with other ecological management systems to promote collaborative prevention and control of regional ecological security. Future development plans for the ecological risk early warning unit include introducing IoT technology, deploying intelligent sensors for soil moisture, water quality parameters, etc., in key areas to achieve real-time monitoring of ecological parameters. The application of deep learning technology in risk identification will be explored to improve early warning accuracy. A comprehensive early warning response plan library will be developed to pre-plan responses for different types of risks. Strengthen cross-regional cooperation and establish a joint early warning mechanism between upstream and downstream areas of river basins. Promote the development of an early warning standard system and participate in the formulation of national standards for ecological risk early warning.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic evaluation system for the effectiveness of land ecological protection and restoration, characterized in that, include: The ecological data acquisition unit is used to acquire land use type data, vegetation cover data, soil physicochemical property data and hydrogeological data of the target area, and integrate them into an ecological basic database. The dynamic feature extraction unit is used to perform spatiotemporal correlation analysis on various types of data in the ecological basic database and extract a set of dynamic feature indicators that reflect the changing trends of land ecological status. The restoration effect modeling unit is used to construct a multi-scale evolution model of the land ecological restoration process based on a dynamic feature index set, and generate a sequence of restoration effect evaluation parameters. Anomaly identification unit is used to establish an ecological state deviation detection mechanism based on the restoration effect evaluation parameter sequence and output a spatial distribution map of land ecological anomaly areas.

2. The dynamic evaluation system for land ecological protection and restoration effects according to claim 1, characterized in that, The specific method by which the ecological data acquisition unit obtains vegetation coverage data is as follows: Vegetation indices were retrieved from multispectral remote sensing images, and a pixel-scale correction model was established by combining ground-measured quadrat data. The spatial continuity of the corrected vegetation index data was reconstructed using spatiotemporal kriging interpolation. Based on the reconstruction results, vegetation cover levels are divided and the area proportion of each level is calculated to form a time-series variation matrix of vegetation cover data.

3. The dynamic evaluation system for land ecological protection and restoration effects according to claim 1, characterized in that, The dynamic feature extraction unit performs the following during spatiotemporal correlation analysis: Spatial autocorrelation models were established for pH value, organic matter content and heavy metal concentration in soil physicochemical properties data; Detect the lag correlation coefficient between groundwater level fluctuations and surface runoff in hydrogeological data; Land use type data are converted into landscape pattern indices, and their coupling and coordination degree with vegetation cover data is calculated.

4. The dynamic evaluation system for land ecological protection and restoration effects according to claim 3, characterized in that, The process of generating the dynamic feature index set includes: Soil parameters with Moran index greater than the threshold in the spatial autocorrelation model were selected as key soil features; Screening hydrological interaction features with lag correlation coefficients exceeding the significance level; Extract landscape-vegetation synergistic variation features with continuously decreasing coupling coordination. Key soil characteristics, hydrological interaction characteristics, and landscape-vegetation synergistic variation characteristics are spliced ​​together into a dynamic feature index set according to time steps.

5. The dynamic evaluation system for land ecological protection and restoration effects according to claim 1, characterized in that, When the repair effect modeling unit constructs a multi-scale evolution model: At the pixel scale, a random forest algorithm is used to predict vegetation restoration potential; Construct coupled differential equations for soil erosion and nutrient cycling at the watershed scale; Establish a game equilibrium model of ecological carrying capacity and restoration investment at the regional scale.

6. The dynamic evaluation system for land ecological protection and restoration effects according to claim 5, characterized in that, The method for generating the repair effect evaluation parameter sequence is as follows: Spatial clustering is performed on the pixel-scale prediction results, and the recovery potential value of the cluster center points is extracted; The steady-state solution of the coupled differential equation is used as a parameter for the ecological balance of the watershed; Record the iterative trajectory of the Pareto optimal solution in the game equilibrium model; The three types of parameters are aligned and integrated according to the repair project stage.

7. The dynamic evaluation system for land ecological protection and restoration effects according to claim 1, characterized in that, The process of establishing the ecological state deviation detection mechanism includes: Calculate the sliding window coefficient of variation for each indicator in the parameter sequence for evaluating the repair effect; Identify the set of spatial locations corresponding to periods of sudden increases in the coefficient of variation; Construct an adjacency matrix that reflects the propagation path of abnormal ecological parameters; The core anomaly propagation nodes are determined based on the eigenvectors of the adjacency matrix.

8. The dynamic evaluation system for land ecological protection and restoration effects according to claim 7, characterized in that, The steps for generating the spatial distribution map are as follows: Use the core anomaly propagation node as a seed point for region growth. Anomaly levels were classified using a Dirichlet process hybrid model. A spatial topology map of the abnormal area is generated by overlaying historical restoration project boundary data; Mark the deviation type of the dominant ecological parameter corresponding to each abnormal area.

9. The dynamic evaluation system for land ecological protection and restoration effects according to claim 1, characterized in that, It also includes an ecological risk early warning unit, which operates as follows: The ecological risk index is calculated based on the expansion rate of anomalous regions in the spatial distribution map. When the ecological risk index exceeds a preset threshold, a tiered early warning signal is triggered. Automatically generate early warning reports that include the coordinates of priority areas and the degree of parameter anomalies.

10. The dynamic evaluation system for land ecological protection and restoration effects according to claim 9, characterized in that, The triggering conditions for the graded early warning signal include: The ecological risk index of the core abnormal diffusion node corresponding to the Level 1 warning has increased for three consecutive periods. A Level II warning requires that the area of ​​abnormal regions exceeds a set proportion of the total area of ​​the watershed; The Level 3 warning is activated when the deviation of the dominant ecological parameters includes both soil heavy metal contamination and vegetation degradation.

Images