Spatiotemporal dynamic simulation and evaluation method and system for watershed water ecological products
By constructing space-time dynamic diagrams and hydrological and water ecological models, identifying key nodes and correcting parameters, the problem that existing technology is difficult to fully reflect the multidimensionality and complexity of water ecological products is solved, and efficient and accurate space-time dynamic simulation evaluation of water ecological products is achieved. .
Patent Information
- Application Number
- CN202411189659.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-28
AI Technical Summary
The existing watershed water ecological product evaluation model is difficult to fully reflect the multidimensionality and complexity of water ecological products, especially in simulating the complex interactions between hydrological processes, ecological processes and human activities, and there is no problem with scale effect and scale conversion. Be effectively resolved.
The spatial and temporal dynamic simulation evaluation method of water basin water ecological products is adopted. By determining the research area, acquiring and preprocessing multi-source data, the spatial and temporal dynamic diagram and hydrological water ecological model are constructed, the spatial and temporal correlation is calculated, the key nodes are identified, the model parameters are corrected, and the model parameters are not performed. Deterministic analysis to achieve comprehensive simulation and evaluation of the value of aquatic ecological products.
This method can more accurately reflect the multidimensionality and complexity of water ecological products, improve the efficiency and accuracy of simulation evaluation, and provide technical support for the realization of the value of water ecological products.
Smart Images

Figure CN118709123B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to hydrological and water ecological technology, in particular to a method and system for spatiotemporal dynamic simulation and evaluation of watershed water ecological products. Background Art
[0002] Rivers and lakes not only play a role in maintaining natural ecological functions, but also provide material conditions and service value for the sustainable development of human society. Exploring and transforming the value of river and lake water ecological products is an important means to resolve the contradiction between river and lake ecological protection and economic and social development. The spatiotemporal dynamic simulation and evaluation method of watershed water ecological products is a hot research topic in the intersection of ecology, hydrology and environmental economics. With the intensification of global climate change and the increasing intensity of human activities, watershed ecosystems are facing unprecedented pressure. Accurately evaluating the spatiotemporal dynamic changes of water ecological products is of great significance for achieving sustainable management of watersheds and the construction of ecological civilization. Water ecological products, as various benefits provided by watershed ecosystems to humans, include water regulation, water quality purification, biodiversity maintenance, carbon fixation and regulation, and cultural services, which are important guarantees for the health of watershed ecosystems and human well-being. Accurately simulating and evaluating the spatiotemporal dynamic changes of these products not only helps to deeply understand the functions and values of watershed ecosystems, but also provides an important basis for formulating scientific water resources management policies, implementing effective ecological compensation mechanisms, and optimizing land use planning.
[0003] At present, domestic and foreign scholars have carried out a lot of research on the assessment of watershed water ecological products. Traditional methods are mainly based on static assessment models, which can quantify the spatial distribution of ecosystem services to a certain extent. In recent years, with the development of remote sensing and geographic information system technology, dynamic assessment methods combined with hydrological models have gradually become a research hotspot. For example, the coupling application of SWAT and ecosystem service assessment models has realized the dynamic simulation of water regulation and water purification services. In addition, machine learning-based methods, such as random forests and deep learning models, have also made some progress in predicting the spatiotemporal distribution of ecosystem services.
[0004] However, there are still some technical problems that need to be solved in existing research. Most of the existing models target a single or a few ecosystem services, and it is difficult to fully reflect the multidimensionality and complexity of water ecological products. In particular, there are deficiencies in simulating the complex interactions between hydrological processes, ecological processes and human activities. Scale effects and scale conversion issues are still a bottleneck. There are often significant differences in the assessment results at different spatial and temporal scales, and how to achieve consistent assessment and interpretation of results across scales needs further study. Therefore, research and innovation are needed. Summary of the invention
[0005] The purpose of the invention is to provide a method and system for spatiotemporal dynamic simulation and evaluation of water ecological products in a river basin, in order to solve one of the above problems.
[0006] Technical solution, according to one aspect of the present application, a spatiotemporal dynamic simulation and evaluation method for watershed water ecological products comprises the following steps:
[0007] Step S1, determine the study area and obtain multi-source data, pre-process the multi-source data, form a pre-processed data set and fuse it, build a spatiotemporal dynamic map and a hydrological and water ecological simulation model for the study area based on the fused pre-processed data set, calculate the spatiotemporal correlation of the spatiotemporal dynamic map, and form a spatiotemporal correlation matrix;
[0008] Step S2, read the spatiotemporal dynamic graph of the watershed, calculate the multidimensional importance index of each node in the graph, identify and extract key nodes at different spatial scales based on the node importance and graph structure characteristics, and consider the multi-scale characteristics of the watershed;
[0009] Step S3, retrieve the measured data of key nodes, modify the spatiotemporal dynamic correlation matrix and the parameters of the hydrological and water ecological model through the measured data, and perform uncertainty analysis;
[0010] Step S4: deploy the hydrological and water ecological model with updated parameters, and simulate and evaluate the value of water ecological products by combining the measured data of key nodes and the spatiotemporal dynamic association matrix with updated parameters, and output the simulation evaluation results.
[0011] According to another aspect of the present application, a spatiotemporal dynamic simulation and evaluation system for watershed water ecological products is also provided, comprising:
[0012] at least one processor; and,
[0013] a memory communicatively connected to at least one of the processors; wherein,
[0014] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products described in any of the above technical solutions.
[0015] Beneficial effects can fully reflect the multidimensionality and complexity of water ecological products, improve the efficiency and accuracy of simulation evaluation, and provide technical support for the realization of the value of water ecological products. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a flow chart of the present invention.
[0017] Figure 2 It is a flow chart of step S1 of the present invention.
[0018] Figure 3 It is a flow chart of step S2 of the present invention.
[0019] Figure 4 It is a flow chart of step S3 of the present invention.
[0020] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION
[0021] like Figure 1 As shown, according to one aspect of the present application, a spatiotemporal dynamic simulation and evaluation method for watershed water ecological products comprises the following steps:
[0022] Step S1, determine the study area and obtain multi-source data, pre-process the multi-source data, form a pre-processed data set and fuse it, build a spatiotemporal dynamic map and a hydrological and water ecological simulation model for the study area based on the fused pre-processed data set, calculate the spatiotemporal correlation of the spatiotemporal dynamic map, and form a spatiotemporal correlation matrix;
[0023] Step S2, read the spatiotemporal dynamic graph of the watershed, calculate the multidimensional importance index of each node in the graph, identify and extract key nodes at different spatial scales based on the node importance and graph structure characteristics, and consider the multi-scale characteristics of the watershed;
[0024] Step S3, retrieve the measured data of key nodes, modify the spatiotemporal dynamic correlation matrix and the parameters of the hydrological and water ecological model through the measured data, and perform uncertainty analysis;
[0025] Step S4: deploy the hydrological and water ecological model with updated parameters, and simulate and evaluate the value of water ecological products by combining the measured data of key nodes and the spatiotemporal dynamic association matrix with updated parameters, and output the simulation evaluation results.
[0026] According to one aspect of the present application, step S1 specifically comprises:
[0027] Step S11, determining a research area, acquiring and preprocessing multi-source data to obtain preprocessed multi-source data; fusing the preprocessed multi-source data through a data fusion method based on graph embedding;
[0028] Step S12: constructing a spatiotemporal dynamic map of the watershed based on the preprocessed multi-source data;
[0029] Step S13: establishing a hydrological and water ecological simulation model and obtaining simulation results for the constructed spatiotemporal dynamic diagram of the watershed;
[0030] Step S14: Based on the simulation results and the spatiotemporal dynamic diagram of the watershed, perform spatiotemporal correlation analysis and construct a comprehensive spatiotemporal correlation matrix.
[0031] In this embodiment, multi-source data includes remote sensing images, DEM data, hydrological observation data, meteorological data, land use data, vegetation index, water quality parameters and biodiversity survey data. It can be obtained through water level meters, flow meters and water quality monitors. Rainfall, temperature, humidity and wind speed data can be obtained through meteorological stations. Water level data is obtained through monitoring stations. Remote sensing impact data can use Landsat 8 OLI and Sentinel-2 MSI images. In key areas, drones with multispectral cameras are used to obtain relevant data. DEM data can be SRTM DEM data with a resolution of 30 meters. For important areas, such as rivers and important water conservancy project areas, airborne LiDAR is used to obtain elevation data with a resolution of 1 meter. Economic data and ecosystem data are obtained from relevant departments.
[0032] The preprocessing process of the relevant data is as follows: atmospheric correction is performed using the FLAASH module of ENVI software, and local atmospheric parameters and sensor information are input. Based on 20 evenly distributed ground control points, geometric correction is performed using the orthorectification tool of ArcGIS.
[0033] FLAASH module (Fast Line-of-sight Atmospheric Analysis of SpectralHypercubes) is a fast atmospheric line-of-sight spectral analysis module, which is mainly used to perform atmospheric correction on hyperspectral and multispectral images. It can eliminate or reduce the influence of the atmosphere on remote sensing images, such as aerosol scattering, water vapor absorption, etc., thereby improving the accuracy and reliability of remote sensing data.
[0034] Apply adaptive multi-scale wavelet denoising algorithm: Decompose the image into 3 layers using the 'db4' wave basis function. Calculate the local variance of each subband and set the adaptive threshold T = k * σ local / σ global , where k is the empirical coefficient, which is 2. Soft threshold processing is performed on the wavelet coefficients: coef new = sign(coef) * max(|coef| - T, 0). Perform three-layer wavelet reconstruction to obtain the denoised image. Use the MosaicPro module of ERDAS Imagine software to mosaic the images, and use the feathering algorithm to process the seam lines.
[0035] The 'db4' wave basis function is a type of Daubechies wavelet family, where 'db' stands for Daubechies and '4' stands for the wavelet order. The 'db4' wave basis function has better frequency localization capabilities and can more accurately capture the frequency characteristics of the signal. 'db4' has 4 vanishing moments and can accurately represent cubic polynomials. This makes it perform well in compression and denoising, and can effectively preserve the important features of the signal. It provides good time-frequency resolution and strikes a balance between computational efficiency. The 'db4' wave basis function performs well in processing various types of signals, including smooth signals and mutation signals. This makes it widely applicable when processing complex hydrological and ecological data.
[0036] For DEM data, a terrain correction algorithm based on curvature flow is applied:
[0037] Calculate the curv = d²z / dx² + d²z / dy² of each grid. x and y are coordinates. Simulate water flow flow_dir = argmax(z_neighbor - z_center + ε). Iterate to fill the depression z_new = max(z, min(z_neighbor) + δ) until there is no depression. Use local polynomial interpolation with a 5×5 window to smooth the corrected terrain. Note that d here is the symbol of the partial derivative operator.
[0038] Using the ArcGIS Hydrology toolset: Calculate slope and aspect: slope = arctan(sqrt[(dz / dx)² + (dz / dy)²]). Generate flow direction rasters using the D8 algorithm, assigning flow direction values from 1 to 8 to each raster. Calculate cumulative flow, flow_acc = sum(upstream_cells). Extract river networks, with the river threshold set to cumulative flow greater than 1000 rasters. Use the watershed algorithm to divide sub-basins.
[0039] For hydrological observation data, anomaly detection and data interpolation are performed:
[0040] Calculate the corrected Z-score: Z = (x - median) / MAD, where MAD is the median absolute deviation. If |Z|> 3.5, mark it as an outlier.
[0041] For missing data less than 6 hours, cubic spline interpolation is used. For long-term missing data, LSTM neural network is used, and the input includes historical data, periodic characteristics and related meteorological factors.
[0042] Draw a double cumulative curve: ∑P vs. ∑Q. Calculate the change in the slope of the curve, and if the change exceeds 10%, perform a regression breakpoint analysis.
[0043] For meteorological data, the correlation coefficient between each station and the five neighboring stations was calculated. If the correlation coefficient of a station was significantly lower than that of other stations (p < 0.05), it was marked as suspicious data.
[0044] The 3DVAR cost function J(x) is constructed, and the conjugate gradient method is used to minimize J(x) to obtain the optimal analysis field. Considering the influence of terrain, the Barnes interpolation method is used to interpolate the analysis field to a 1km resolution grid.
[0045] The land use data were processed by random forest algorithm, using random forest classifier, and the features included multispectral bands, texture features and terrain index. 500 decision trees were set, and the minimum number of samples for each node was 10. Multiscale segmentation algorithm was applied to merge adjacent pixels with similar spectra. The majority filter was used for post-processing of 3×3 windows to eliminate "salt and pepper" noise.
[0046] Construct a U-Net convolutional neural network with the input being the superposition of two phases of images. Use the focal loss function to solve the class imbalance problem. Set a threshold of 0.5 and convert the output probability map into a binary change map.
[0047] Construct multi-level node representation, macro scale: take hydrological units (such as sub-basins at the HUC12 level) as nodes. Calculate the area, average elevation, and proportion of major land use types for each unit.
[0048] Mesoscale: Identify key hydrological points, including river confluences, reservoirs, water intakes, etc. Extract the latitude and longitude coordinates and related attributes (such as flow, water quality parameters) of each point.
[0049] Microscale: Divide the study area into a regular grid of 1 km × 1 km. Assign a unique identifier and grid coordinates to each grid cell.
[0050] Construct node attribute vectors, including hydrological attributes: calculate annual average flow, flow variation coefficient, peak flow. Extract groundwater level, aquifer thickness, and permeability coefficient. Ecological attributes: calculate NDVI mean and standard deviation, biodiversity index (such as Shannon index). Extract dominant species types and biomass estimates. Socioeconomic attributes: count population density, GDP, and major industry types. Calculate water consumption and pollutant emissions.
[0051] Node time series association: Create a timestamp index for each node and associate historical observation data. Design a sliding window mechanism to dynamically update the node status of the last N time steps.
[0052] Construct a hydrological connectivity matrix, including:
[0053] The flow direction matrix D is calculated based on DEM, where Dij = 1 means there is water flow from node i to node j, otherwise it is 0.
[0054] Calculate the cumulative flow matrix A = (I - D) -1 , where I is the identity matrix.
[0055] Construct the hydrological distance matrix H, Hij = min (cumulative flow value on the path).
[0056] Ecological interaction matrix construction, including:
[0057] Compute the geographic distance matrix G between nodes.
[0058] Construct the species similarity matrix S using the Jaccard index.
[0059] Define the ecological distance matrix E = 1 / (S * exp(-αG)), where α is the attenuation coefficient.
[0060] Socio-economic linkage matrix, including:
[0061] Establish a traffic network graph T, where the edge weight is the path distance.
[0062] Calculate the economic flow matrix F, using the gravity model Fij = (GDPi * GDPj) / Tij β .
[0063] Comprehensive edge weight calculation, including:
[0064] Construct a multi-objective optimization problem: min w1f(H) + w2f(E) + w3*f(F); st w1 + w2 + w3 =1, wi ≥ 0.
[0065] Use the NSGA-II algorithm to find the Pareto optimal solution set. Select a balanced solution as the final edge weight.
[0066] Event-driven graph structure updates:
[0067] Monitor the real-time data stream to detect whether the trigger conditions are met. If the conditions are met, update the attribute vector of the affected node and recalculate the weights of the related edges.
[0068] Recursive Bayesian estimation of smoothers:
[0069] State equation: x(t) = f(x(t-1)) + w(t), where w(t) is process noise. Observation equation: y(t) = h(x(t)) + v(t), where v(t) is observation noise. Prediction step: x̂(t|t-1) = f(x̂(t-1|t-1)). Update step: x̂(t|t) = x̂(t|t-1) + K(t)[y(t) - h(x̂(t|t-1))]. Where K(t) is the Kalman gain.
[0070] The specific process of graph compression is as follows:
[0071] Identify key nodes in the network. Start by calculating three centrality metrics: degree centrality, betweenness centrality, and eigenvector centrality. Degree centrality reflects the number of direct connections a node has, betweenness centrality measures the importance of a node as a bridge for information transfer in the network, and eigenvector centrality takes into account the importance of a node's neighbors.
[0072] Assess the hydro-ecological importance of each node. Construct a comprehensive importance index that not only includes the centrality index calculated above, but also incorporates specific hydrological and ecological importance indicators. This index combines the various indicators together in a weighted sum, and the choice of weights reflects the relative importance of different factors in the assessment.
[0073] Perform iterative graph simplification. A target number of nodes and an importance threshold are set as stopping conditions. In each iteration, the node with the least importance is found and merged with its most important neighbor node. This merging process needs to be handled carefully to ensure that the overall structure and properties of the network do not change significantly. A new node is created to replace the merged node, and this new node inherits all the connections of the original node. For numerical attributes of the node, a weighted average is used to calculate the new value; for categorical attributes, a weighted majority method is used. The weights of the edges connected to the new node also need to be updated accordingly, also using the weighted average method.
[0074] After each merge operation, the global topological features of the graph, such as average path length and clustering coefficient, are recalculated. If the change in these features exceeds a preset threshold (e.g. 5%), the merge operation is rolled back to ensure that the compression process does not excessively change the basic structure of the network.
[0075] To ensure that the compressed graph still retains the key features of the original graph, a consistency check is performed. This check is mainly based on the spectral distribution of the graph, that is, the distribution of the eigenvalues of the Laplacian matrix. The first k non-zero eigenvalues of the original and compressed graphs are calculated, and then the Kolmogorov-Smirnov test is used to compare the two distributions. If the p-value of the test is less than 0.05, it means that the compression may be over-compression, so the target number of nodes is increased and the compression process is re-executed.
[0076] Through the graph compression process, the complexity of the watershed network can be significantly reduced while preserving its key structural and functional characteristics.
[0077] Specifically, the graph compression process includes: Key node identification: Calculate the degree centrality, betweenness centrality, and eigenvector centrality of nodes.
[0078] The process of hydrological and ecological importance assessment includes: Defining the node importance index: I(v) = w1CD(v) + w2CB(v) + w3CE(v) + w4H(v) + w5*E(v). Where H(v) and E(v) are hydrological and ecological importance indicators respectively.
[0079] Set the target number of nodes N' and the importance threshold T. When the number of nodes > N' and the minimum importance < T: Find the node v with the minimum importance. Identify the neighbor node set U of v. Merge v with the node with the maximum importance in U.
[0080] Create a new node w, inheriting all the edges of the merged nodes. For the new node, calculate the attributes: For numerical attributes: xw = (xvIv + xuIu) / (Iv + Iu); For categorical attributes: Use the weighted mode; Update the weights of the edges connected to w: wnew = (wvIv + wuIu) / (Iv + Iu).
[0081] Recalculate the global topological characteristics of the graph, such as the average path length and clustering coefficient. If the global characteristics change by more than a preset threshold (such as 5%), roll back this merge operation.
[0082] Calculate the spectral distributions of the original graph and the compressed graph: Solve the eigenvalue problem Lx = λx, where L is the Laplacian matrix. Calculate the first k non-zero eigenvalues: λ1 ≤ λ2 ≤... ≤ λk. Use the Kolmogorov-Smirnov test to compare the two spectral distributions. If the p-value < 0.05, increase the target number of nodes N' and re-execute the compression.
[0083] Step S13: Construction of the hydrological and hydro-ecological simulation model
[0084] Based on the constructed watershed spatio-temporal dynamic graph, establish a modular and multi-scale hydrological and hydro-ecological coupling model. Perform parameter initialization, model structure optimization, and uncertainty quantification to provide a reliable model basis for subsequent simulations and evaluations.
[0085] The process of initializing the parameters of the target watershed includes:
[0086] Extract the feature vector f_target of the target watershed.
[0087] Use the meta-learner to predict the initial parameters: θ_init = RF(f_target).
[0088] Define parameter prior distribution: p(θ) ~ N(θ_init, Σ), where Σ is a predefined covariance matrix.
[0089] The process of Bayesian parameter fine-tuning includes:
[0090] Construct the likelihood function and posterior distribution, use the Metropolis-Hastings algorithm to sample the posterior distribution, and take the posterior mean as the final parameter estimate.
[0091] The process of optimizing the model structure includes:
[0092] Construct a candidate model set: M = {M1, M2, ..., Mk}, where each Mi represents a sub-model combination.
[0093] Design of multi-objective evaluation indicators: Simulation accuracy: Nash-Sutcliffe efficiency coefficient (NSE). Computational efficiency: average running time t. Model complexity: number of parameters n.
[0094] The process of model structure adjustment includes:
[0095] Based on the optimal model, make small structural changes (such as adding / deleting a submodule). Use parameter sensitivity analysis to identify key parameters and redundant parameters. Apply model simplification techniques, such as Akaike Information Criterion (AIC), to balance model complexity and goodness of fit.
[0096] Parameter uncertainty, construct parameter posterior probability density function (PDF): Use DREAM (DiffeRential Evolution Adaptive Metropolis) algorithm for MCMC sampling. Generate parameter posterior sample set.
[0097] Compute parameter confidence intervals and, for each parameter, calculate the 2.5% and 97.5% quantiles.
[0098] Structural uncertainty, implement Bayesian model averaging (BMA), and calculate model weights.
[0099] Scenario uncertainty, construct key scenario factors (such as climate change, land use change). Generate scenario combinations using orthogonal experimental design. Perform Monte Carlo simulations for each scenario.
[0100] The process of comprehensive uncertainty analysis is as follows:
[0101] Variance decomposition: The contribution of each uncertainty source is calculated using the Sobol method; for each time step, the 2.5% and 97.5% quantiles of the simulation results are calculated.
[0102] Calculate the frequency of the observed value in the prediction interval and use CRPS (Continuous Ranked Probability Score) to evaluate the probability forecast skill.
[0103] Step S14: Based on the constructed spatiotemporal dynamic diagram and simulation model of the watershed, the spatiotemporal correlations among hydrological, ecological and socio-economic elements are deeply analyzed to provide a theoretical basis for the evaluation of water ecological products.
[0104] Multi-scale spatiotemporal correlation analysis:
[0105] Wavelet decomposition of time series: Select Morlet wavelet as the mother wavelet function. Perform continuous wavelet transform on the time series of each node and calculate the wavelet power spectrum.
[0106] For each pair of nodes (i, j), the cross-wavelet coherence is calculated, and at each scale a, the average coherence is calculated.
[0107] Construct the spatial weight matrix W based on inverse distance weighting. Calculate the global Moran's I index and the local Moran's I index.
[0108] Construct a spatiotemporal adjacency matrix A, where nodes i and j are adjacent in space and time steps k and l are adjacent. Compute the graph: solve the generalized eigenvalue problem, select the first k eigenvectors, and construct a low-dimensional representation.
[0109] For each pair of time series (X, Y), fit the VAR model and perform an F test. If the p-value is < 0.05, then X is considered to Granger cause Y.
[0110] Estimate joint probability distribution, calculate transfer entropy and conditional mutual information, estimate high-dimensional probability density using k-nearest neighbor method, and calculate conditional mutual information.
[0111] Construct a causal network: Initialize a fully connected undirected graph G. For each pair of nodes (X,Y): If TE(X→Y) >TE(Y→X), add a directed edge X→Y. Otherwise, add a directed edge Y→X.
[0112] For each triple (X,Y,Z): if I(X;Y|Z) < ε, remove the XY edge. Apply FCI and apply the FCI (FastCausal Inference) algorithm for causal skeleton learning. Specifically: perform conditional independence tests and gradually remove redundant edges. Identify V-shaped structures. Infer the direction of the edges and obtain a partial directed acyclic graph.
[0113] Calculate the causal strength CS of each edge, and remove all edges with CS < τ according to the set threshold τ. Use the transfer closure algorithm to detect and eliminate loops.
[0114] For each retained edge X→Y, fit an additive model. Use generalized cross validation (GCV) to select the optimal number of spline basis functions. Compute the average causal effect.
[0115] The process of moment pattern extraction is as follows:
[0116] Multidimensional spatiotemporal data are obtained, and causal inference methods are used to analyze the relationship between variables. First, an additive model is established for each pair of related variables, which includes the target variable, the main explanatory variable and other influencing factors. Then, the generalized cross-validation method is used to optimize the number of spline basis functions in the model to balance the model's fit and complexity. Finally, the average causal effect is calculated to quantify the degree of influence of the main explanatory variables on the target variable, thereby revealing the strength of the causal relationship between the variables.
[0117] A multidimensional tensor is constructed to represent spatiotemporal data, and the Tucker decomposition method is used to extract data features. First, the data is organized into fourth-order tensors, representing spatial position, time step, variable type, and feature dimension respectively. In order to enhance the local structure of the time dimension, the sliding window method is used to process time series data. Then, the objective function of Tucker decomposition is defined, and the alternating least squares algorithm is used to iteratively solve the core tensor and factor matrix. Finally, the Bayesian information criterion is used to select the optimal rank combination to achieve a balance between model complexity and fitting accuracy.
[0118] The decomposed data is sparsely encoded, and the extracted patterns are clustered and visualized. First, a dictionary and sparse coefficients are constructed for each data pattern, and the online dictionary learning algorithm is used to solve it, and a threshold is set to filter important coefficients. Then, the cosine similarity of the extracted patterns is calculated, and the Ward method is used for hierarchical clustering, and the optimal number of clusters is determined by calculating the silhouette coefficient. Finally, the t-SNE algorithm is used to map the high-dimensional pattern to a two-dimensional plane, intuitively displaying the distribution characteristics of the data pattern, which is convenient for further analysis and interpretation.
[0119] The process of building the correlation matrix is as follows:
[0120] The geographic coordinate data of the study area was obtained, and the spatial analysis method was used to calculate the proximity between regions. First, the connection relationship between spatial points was constructed using the Delaunay triangulation algorithm. Then, important spatial connections were screened based on the relative proximity graph theory to ensure that only the most representative proximity relationships were retained. Finally, the Gaussian kernel function was used to convert the spatial distance into a continuous proximity index to obtain a quantitative index reflecting the regional spatial relationship.
[0121] The hydrological network data of the study area were obtained and the hydrological connectivity index was calculated. First, a directed graph representing the topological structure of the river system was constructed. Then, the upstream contribution area of each node was obtained through recursive calculation, reflecting its importance in the hydrological system. Finally, the hydrological connectivity was defined based on the common upstream area, and an index was obtained to quantify the strength of the hydrological relationship between different locations.
[0122] The ecological and environmental data of the study area were obtained to evaluate the ecological similarity between regions. First, multidimensional ecological indicators including vegetation index and biodiversity were collected. Then, the Euclidean distance between the ecological indicator vectors of different locations was calculated to quantify the degree of difference in their ecological characteristics. Finally, the Gaussian kernel function was used to convert the distance into a similarity index to obtain a quantitative index reflecting the degree of similarity of the ecological environment between regions.
[0123] Obtain socioeconomic data for the study area and evaluate the strength of socioeconomic connections between regions. First, construct a weighted graph reflecting the transportation network, where edge weights represent travel time. Then, use the Floyd-Warshall algorithm to calculate the shortest path distance between regions. Finally, based on the principle of the gravity model, combined with factors such as population and GDP, calculate the socioeconomic connection strength index between regions.
[0124] Integrate multi-dimensional correlation indicators and calculate the comprehensive correlation. First, construct a multi-objective optimization problem, the goal is to maximize the correlation between the comprehensive correlation and each single indicator. Then, use the non-dominated sorting genetic algorithm to solve the Pareto optimal solution set. Finally, use the TOPSIS method to select the best weight combination from the Pareto solution set to obtain a comprehensive correlation index that balances the influence of various factors.
[0125] The correlation matrix is constructed and optimized based on the comprehensive correlation. First, the initial correlation matrix is calculated using the optimal weight combination. Then, the spectral clustering method is applied to identify the community structure in the network. Then, the correlation strength is adjusted according to the community structure to strengthen the internal connection of the community. After that, the PageRank algorithm is used to calculate the importance of the node. Finally, the correlation strength is adjusted again based on the importance of the node to obtain the optimized correlation matrix reflecting the interaction of the complex geographical system.
[0126] This embodiment achieves comprehensive characterization and efficient simulation of the water ecosystem in the watershed through multi-source data acquisition and preprocessing, construction of a spatiotemporal dynamic map of the watershed, and construction of a hydrological and water ecological simulation model. Specifically, an adaptive multi-scale wavelet decomposition algorithm is used to denoise and enhance remote sensing images, which can automatically adjust the decomposition scale and threshold according to the local features of the image, and retain edge information while improving image quality, which is crucial for accurately identifying water body boundaries and vegetation coverage changes. A terrain correction algorithm based on curvature flow is applied to DEM data. By simulating the water flow movement process, low-lying areas are adaptively filled and abnormal high points are eliminated to generate a terrain model with better hydrological consistency, thereby improving the accuracy of subsequent hydrological simulations. The developed multi-source data fusion method based on graph embedding represents various types of heterogeneous data (such as remote sensing images, hydrological observations, socio-economic statistics, etc.) as a graph structure, learns a low-dimensional representation of each data through a graph convolutional network, and then uses a multi-head attention mechanism to adaptively fuse these representations to obtain a unified data representation. This method not only solves the problem that traditional data fusion methods are difficult to handle high-dimensional heterogeneous data, but also captures the complex nonlinear relationship between data, providing a more comprehensive and accurate data basis for subsequent water ecological product evaluation. In terms of spatiotemporal interpolation, the proposed adaptive Kriging interpolation method combining terrain characteristics and hydrological processes significantly improves the interpolation accuracy by introducing physically based covariates (such as slope, runoff accumulation, etc.) and dynamically adjusting the variogram parameters, especially in data sparse areas. This is of great significance for evaluating the spatial distribution of water ecological products in large-scale watersheds. In the process of constructing the spatiotemporal dynamic graph of the watershed, the designed multi-level and multi-scale node representation method combines hydrological units, key hydrological points and grid units to achieve a comprehensive characterization from macro to micro. The adaptive edge weight calculation method based on hydrological connectivity and ecological interaction balances different types of connection strength by solving multi-objective optimization problems, accurately depicting the complex interactive relationship between hydrological and ecological processes in the watershed. The developed event-triggered dynamic graph update mechanism can respond to sudden changes in the watershed system (such as extreme rainfall events) in a timely manner, ensuring the sensitivity and adaptability of the model to actual conditions. The graph compression algorithm simplifies the graph structure through an iterative merging process while maintaining the hydrological process and ecological functions, greatly reducing the computational complexity and making long-term simulation of large-scale watersheds possible. In terms of the construction of hydrological and water-ecological simulation models, the proposed modular, multi-scale coupled model framework includes five main modules: hydrology, hydraulics, water quality, ecology, and socio-economics, achieving a comprehensive integration of all elements of the watershed system. The parameter initialization method based on transfer learning uses the knowledge of existing research basins to significantly accelerate the adaptation process of the model in new basins. The adaptive model structure optimization algorithm improves the flexibility and applicability of the model by dynamically selecting the model structure that best suits the current basin characteristics and simulation objectives.The parameter uncertainty quantification method based on information geometry uses the Fisher information matrix to define the distance metric and provides more accurate uncertainty estimates through MCMC sampling, which is of great significance for evaluating the reliability of the value of water ecological products.
[0127] According to one aspect of the present application, step S2 is specifically:
[0128] Step S21, read the node attribute data in the spatiotemporal dynamic graph, calculate the importance indexes of the three dimensions of hydrology, ecology and socio-economy, use the hierarchical analysis method to determine the weight of each index, and comprehensively obtain the multidimensional importance index of each node;
[0129] Step S22, sort all nodes in descending order based on the multidimensional importance index, and select the first N nodes as the initial candidate set; calculate the graph topology features of the nodes; construct the node importance score, combining the multidimensional importance index and the graph topology features; use the greedy algorithm to iteratively select the node with the highest score, and update the scores of the remaining nodes after each selection; calculate the spatial distribution index of the selected key node, and if it is lower than the threshold, adjust the node selection while maintaining the overall importance; N is a natural number;
[0130] Step S23, constructing a spatial scale sequence; constructing a corresponding graph structure for each scale, merging nodes in the same grid into super nodes; calculating the characteristics of the graph structure at each scale, and the characteristic change rate between adjacent scales; using the silhouette coefficient method to select the optimal scale; applying the key node identification algorithm on the optimal scale; for the key super nodes containing multiple original nodes, recursively applying the identification algorithm; merging the key nodes of all levels to obtain the final multi-scale key node set;
[0131] Step S24, divide the total simulation time into several overlapping time windows, and for each time window, calculate the time series characteristics of the nodes; update the node importance score and reapply the key node identification algorithm; calculate the Jaccard similarity of the key node sets in adjacent time windows, and if the similarity is lower than the threshold, perform smoothing adjustment; analyze the frequency of nodes being selected as key nodes in all time windows, classify the nodes into stable key nodes, periodic key nodes and instantaneous key nodes, and use the time series prediction model to predict the potential key nodes of the next time window.
[0132] In this embodiment, the importance index of each node in the hydrological, ecological and socio-economic dimensions is calculated to comprehensively evaluate the overall importance of the node.
[0133] Steps S21 to S24 are specifically as follows:
[0134] Step S211, calculate the hydrological importance index of the basin nodes. First, determine the Strahler level of each node based on the river network structure to reflect its position in the river hierarchy. The specific method is to initialize the levels of all source nodes to 1, and then recursively calculate downstream along the river: if the levels of all upstream nodes of a node are the same, the level of the node is the upstream level plus 1; otherwise, take the maximum level among the upstream nodes. Then, calculate the cumulative flow contribution of each node, that is, the node's own catchment area plus the cumulative flow of all upstream nodes. Next, evaluate the flood storage capacity of the node, which is expressed by the ratio of the node's storage capacity to the difference between the maximum inflow and outflow. Finally, the three indicators are normalized and weighted summed to obtain a comprehensive hydrological importance index.
[0135] Step S212, evaluate the ecological importance index of the watershed nodes. First, calculate the biodiversity index of the node, and you can choose the Shannon index or the Simpson index. The Shannon index takes into account the richness and uniformity of species, and is obtained by the logarithmic operation of the relative abundance of species; the Simpson index reflects the concentration of dominant species in the community. Then, estimate the ecosystem service value of the node by multiplying the area of different land use types by the corresponding value coefficient and summing them. Next, evaluate the ecological fragility of the node, considering multiple influencing factors such as slope, vegetation coverage, etc., and assigning different weights. Finally, the biodiversity index, ecosystem service value and ecological fragility (negation) are normalized and weighted to obtain a comprehensive ecological importance index.
[0136] Step S213, calculate the socio-economic importance index of the basin node. First, calculate the population density of the node, that is, the total population within the node divided by the area. Then, calculate the economic density of the node, that is, the total GDP within the node divided by the area, to reflect the concentration of economic activities. Next, evaluate the water resource utilization intensity of the node, expressed by the ratio of annual water consumption to annual available water, to reflect the pressure of water resource utilization. Finally, the three indicators of population density, economic density and water resource utilization intensity are normalized and weighted summed to obtain a comprehensive socio-economic importance index.
[0137] Step S214, integrate the importance indexes of the three dimensions of hydrology, ecology and socio-economics, and calculate the comprehensive importance index of the node. First, use the analytic hierarchy process (AHP) to determine the weights of the three dimensional indexes. The specific steps include: constructing a judgment matrix, in which the matrix elements represent the relative importance of different indicators; calculating the maximum eigenvalue and the corresponding eigenvector of the judgment matrix to obtain a preliminary weight vector; performing a consistency test to ensure the rationality of the judgment. If the consistency ratio is less than 0.1, the weight vector is accepted; otherwise, the judgment matrix needs to be reconstructed. Finally, using the determined weights, the hydrological importance index, ecological importance index and socio-economic importance index are weighted and summed to obtain the final index reflecting the comprehensive importance of the node.
[0138] Step S221, generate an initial candidate node set. First, sort all nodes in descending order of comprehensive importance index. Then, select a certain proportion (such as 10%) of the top ranked nodes as the initial candidate set. This step ensures that nodes with higher comprehensive importance are given priority.
[0139] Step S222, analyze the graph topology characteristics of the nodes. First, calculate the degree centrality of each node, that is, the ratio of the number of adjacent nodes of the node to the total number of nodes, reflecting the direct influence of the node. Then, calculate the betweenness centrality of the node, which indicates the degree to which the node acts as a bridge in the network, and is evaluated by calculating the number of shortest paths passing through the node. Next, calculate the closeness centrality of the node, which reflects the average distance from the node to all other nodes, indicating the efficiency of information dissemination. Finally, calculate the eigenvector centrality of the node, considering the importance of the node's neighbors, and obtain it by solving the characteristic equation of the adjacency matrix.
[0140] Step S223, select key nodes. First, define the node importance score, and perform a weighted combination of the comprehensive importance index and various centrality indicators. Then, use a greedy algorithm to gradually select the nodes with the highest scores as key nodes. After each node is selected, update the scores of the remaining nodes, consider the impact of the selected nodes on other nodes, and avoid selecting nodes that are too similar. Repeat this process until a preset number of key nodes are selected.
[0141] Step S224, optimize the spatial distribution of key nodes. First, calculate the spatial distribution index of the selected key nodes to reflect the degree of dispersion of the nodes in the geographic space. Then, if the spatial distribution index is lower than the preset threshold, perform adjustments: find a pair of key nodes with the closest geographical distance, search for non-key nodes with higher scores around them, and use this node to replace the one with the lower score in the original pair. Repeat this process until the spatial distribution index meets the requirement or the maximum number of iterations is reached.
[0142] Step S231, construct a multi-scale graph. First, define a series of spatial scales, such as 1km, 5km, 10km, 50km, etc. Then, for each scale, use the grid method to divide the basin into several units, and merge the nodes falling in the same grid into a supernode. Finally, calculate the edge weights between supernodes, which are obtained by merging all connections between original nodes.
[0143] Step S232: extract scale features. First, calculate the basic features of the graph at each scale, including the number of nodes, the number of edges, the average degree, the clustering coefficient, and the average path length. Then, calculate the change rate of these features between adjacent scales to reflect the degree to which the network structure changes with scale.
[0144] Step S233, select the optimal scale. Use the silhouette coefficient method to evaluate the clustering effect of each scale. First, for each node, calculate its average distance to other nodes in the same cluster and its average distance to the nearest other cluster. Then, calculate the silhouette coefficient of the node based on these two distances. Finally, select the scale with the largest average silhouette coefficient as the optimal scale.
[0145] Step S234, adaptively extract key nodes. First, apply the key node identification algorithm in step S22 on the optimal scale. Then, for each identified key super node, if it contains multiple original nodes, recursively apply the key node identification algorithm. Finally, merge the key nodes of all levels to obtain the final multi-scale key node set.
[0146] Step S241, divide the time window. Divide the total simulation time into a series of overlapping time windows, each window has a fixed length, and there is a certain degree of overlap between adjacent windows. This division method can capture the continuous changes of the system and identify sudden events.
[0147] Step S242, calculate the time series characteristics of the nodes in the window. First, calculate the mean and standard deviation of each node in the window to reflect the central tendency and volatility of the node state. Then, use the Mann-Kendall test to calculate the trend coefficient to determine the upward or downward trend of the node state. Finally, analyze the periodicity of the node state through the autocorrelation function and fast Fourier transform.
[0148] Step S243: Update the node importance score. Combine the time series features of the node with the previously calculated importance score to obtain a new score that reflects the dynamic importance of the node. Then, apply the key node identification algorithm in each time window to obtain the key node set of the window.
[0149] Step S244, smoothly update the key node set. First, calculate the Jaccard similarity of the key node sets in adjacent time windows to evaluate the degree of change of the key node sets. If the similarity is lower than the threshold, perform adjustment: the high-scoring nodes added in the new window may replace the low-scoring nodes in the old window. Repeat this process until the similarity meets the requirement or the maximum number of iterations is reached.
[0150] Step S245, analyze the long-term stability of the node. Calculate the frequency of each node being selected as a key node in all time windows. Classify the nodes into stable key nodes, periodic key nodes, and transient key nodes according to the frequency. For periodic key nodes, use wavelet analysis to identify their significant cycles.
[0151] Step S246: Predictively update key nodes. First, use a time series prediction model (such as ARIMA or Prophet) to predict the node status of the next time window. Then, calculate the expected importance score of the node based on the predicted status. Finally, incorporate the expected score into the key node selection process to obtain the predicted key node set for the next time window.
[0152] By extracting key nodes from the spatiotemporal dynamic graph of the watershed, efficient characterization and analysis of complex watershed systems are achieved. In terms of multidimensional node importance assessment, the analytic hierarchy process (AHP) is used to determine the weights of the three dimensional indicators of hydrology, ecology and socio-economics, and the importance index of each node is calculated comprehensively. This method not only considers traditional hydrological indicators (such as Strahler level and cumulative flow contribution), but also introduces ecological indicators (such as biodiversity index, ecosystem service value) and socio-economic indicators (such as population density and water resource utilization intensity), which fully reflects the importance of nodes in the formation and value realization of water ecological products. This multidimensional evaluation method is particularly suitable for complex watershed systems, and can identify key areas that have a significant impact on the value of water ecological products, providing a focus for subsequent refined simulation and management. In the key node identification algorithm, the node importance index and graph topological characteristics (such as degree centrality, betweenness centrality, etc.) are combined, and a greedy algorithm is used for selection. The selection of redundant nodes is avoided by dynamically updating the scores of the remaining nodes. This method not only ensures the importance of the selected nodes, but also ensures the complementarity between nodes, and can capture the main characteristics of the watershed system with the least number of nodes. In particular, in the optimization of spatial distribution, by adjusting the geographical distribution of key nodes, it is ensured that the selected nodes can fully cover all important areas of the basin, which is of great significance for accurately evaluating the spatial heterogeneity of water ecological products. In terms of multi-scale key node extraction, a scale adaptive algorithm is innovatively introduced. By constructing a graph structure at different spatial scales (such as 1km, 5km, 10km and 50km) and using the silhouette coefficient method to select the optimal scale, the problem that the traditional single-scale method is difficult to adapt to basins of different sizes is solved. This multi-scale method is particularly suitable for the evaluation of water ecological products in large basins, and can capture local detail features while ensuring the overall characterization accuracy. For key super nodes containing multiple original nodes, the strategy of recursively applying the identification algorithm ensures that the key nodes selected at the coarse scale can accurately reflect the fine-scale information and effectively avoid information loss during the scale conversion process. In terms of the dynamic update of the key node set, the time window sliding and node importance dynamic evaluation mechanism are innovatively introduced. By calculating the statistical characteristics of the node time series (such as mean, standard deviation, trend coefficient and periodicity) and updating the node importance score, the key node set can be adjusted over time. This dynamic update mechanism is particularly suitable for long-term water ecological product value assessment and can timely capture the dynamic changes of the watershed system, such as seasonal changes, extreme event impacts, etc. Jaccard similarity is introduced to evaluate the changes in key node sets in adjacent time windows, and through the smooth adjustment mechanism, it not only ensures the temporal continuity of the key node set, but also reflects the gradual change process of the system.Finally, by analyzing the frequency of nodes being selected as key nodes, the nodes are classified into stable key nodes, periodic key nodes and instantaneous key nodes. This classification method provides a new perspective for understanding the long-term dynamic characteristics of the watershed system. In particular, for periodic key nodes, using wavelet analysis to identify their significant cycles helps to reveal the law of seasonal or interannual changes in the value of water ecological products. The introduction of the time series prediction model enables the system to predict the potential key nodes in the next time window, providing an important basis for forward-looking water ecological product management and decision-making.
[0153] According to one aspect of the present application, step S3 is specifically:
[0154] Step S31, extract the measured data of key nodes, use the modified Z-score method to detect outliers, mark and process abnormal data; apply spatial consistency test, calculate the correlation coefficient between nodes and adjacent nodes, and mark suspicious data; for short-term missing data, use cubic spline interpolation method to supplement; for long-term missing, use multivariate LSTM model for estimation; use ensemble Kalman filter for data assimilation, fuse observation data and model prediction, and obtain the optimal estimate;
[0155] Step S32: Use the Sobol method to perform global parameter sensitivity analysis, identify key parameters, construct a multi-objective function and apply the NSGA-III algorithm to perform multi-objective parameter optimization to obtain a set of Pareto optimal solutions; use the improved DREAM algorithm to perform Bayesian parameter estimation to obtain the posterior distribution of parameters; compare different model structures based on the BIC criterion and construct a multi-model integrated system;
[0156] Step S33: based on the measured data of key nodes, recalculate the Moran's I index and spatiotemporal variogram, update the spatial autocorrelation structure, construct a dynamic weight function, and adjust the correlation strength according to the node observation residuals; use the GraphicalLasso algorithm to estimate the precision matrix and update the adjacency relationship of the graph; calculate the mutual information and transfer entropy between nodes to capture nonlinear associations; apply sliding window analysis and state space modeling to construct a time-varying association model to obtain a revised spatiotemporal dynamic association matrix;
[0157] Step S34, use the MCMC method to generate parameter posterior samples and calculate the posterior statistical characteristics of the parameters; apply the Bayesian model averaging method to quantify the model structure uncertainty; evaluate the input data uncertainty through perturbation analysis and response surface method; construct the prediction interval and use the residual resampling method to consider heteroskedasticity; perform uncertainty decomposition and attribution, and use ANOVA decomposition and Shapley value analysis to quantify the uncertainty contribution from different sources.
[0158] Steps S31 to S34 are specifically as follows:
[0159] Step S311, establish a multi-source data acquisition system. First, deploy an IoT sensor network at each key node, and install water level meters, flow meters, and water quality multi-parameter meters. Then, set up an automatic weather station to collect meteorological elements such as rainfall, temperature, humidity, and wind speed. Finally, install a groundwater monitoring well, equipped with an automatic water level recorder, to comprehensively monitor groundwater dynamics.
[0160] Step S312: Realize real-time data transmission. Use 4G or 5G network to transmit data in real time to ensure the timeliness of data. At the same time, set up local data cache to prevent data loss caused by network interruption and ensure data continuity and integrity.
[0161] Step S313: Collect historical data. Obtain long-term hydrological observation records from the hydrological department, collect ecological survey data (such as vegetation coverage, species diversity, etc.), and integrate socio-economic statistics (such as population, GDP, water consumption, etc.) to provide rich historical background information for subsequent analysis.
[0162] Step S314: Perform data quality control. First, use the modified Z-score method to detect outliers, and mark data points with an absolute Z score greater than 3.5 as potential outliers. Then, perform a spatial consistency test, calculate the correlation coefficient between each node and its neighboring nodes, and mark data with an absolute value of the correlation coefficient less than 0.6 as suspicious data. Finally, fill in the data, use cubic spline interpolation for short-term missing data (less than 6 hours), and apply a multivariate LSTM model for long-term missing data.
[0163] Step S315, perform data assimilation. Construct an ensemble Kalman filter system, generate an initial ensemble, and then alternate between prediction steps and update steps. After that, calculate and correct the model deviation, and use the exponential smoothing method to update the deviation estimate. Finally, perform covariance expansion, determine the optimal expansion parameter through maximum likelihood estimation, and update the ensemble members to prevent filter divergence.
[0164] Step S316, integrating multi-scale data. First, perform discrete wavelet transform on each time series, select Daubechies wavelet as the basis function, and perform 5-layer decomposition. Then, resample the data of different frequencies, and use interpolation or aggregation methods to achieve scale matching. Finally, use Kalman smoother to fuse data from different sources at the matching scale, and reconstruct the fused time series through inverse wavelet transform.
[0165] Step S321, perform parameter sensitivity analysis. Use the Sobol method to calculate the global sensitivity index, generate a large number of parameter samples, calculate the first-order sensitivity index and the total sensitivity index. Then, use the FAST method for verification, design the parameter search curve, calculate the Fourier transform of the model output, and obtain the sensitivity index.
[0166] Step S322, perform multi-objective parameter optimization. First, define multiple objective functions, including Nash-Sutcliffe efficiency coefficient, percentage deviation and Kling-Gupta efficiency. Then, apply NSGA-III algorithm for optimization, generate a uniformly distributed reference point set, and select individuals using non-dominated sorting and reference point method. Finally, adopt adaptive strategy to dynamically adjust crossover and mutation probabilities to improve the convergence and diversity of the algorithm.
[0167] Step S323, implement Bayesian parameter estimation. First, define the parameter prior distribution based on expert knowledge or literature values. Then, construct the likelihood function, assuming that the error follows a Gaussian distribution. After that, use the DREAM algorithm to sample the posterior distribution, run multiple Markov chains in parallel, and use the jump rate and candidate point differences to generate the proposed distribution. Finally, analyze the posterior distribution, calculate the posterior mean and 95% confidence interval of the parameter, and use the Gelman-Rubin diagnostic to evaluate convergence.
[0168] Step S324, optimize the model structure. First, define the candidate model set and calibrate the parameters of each model. Then, calculate the BIC (Bayesian Information Criterion) of each model. After that, calculate the model weight based on the BIC. Finally, build a multi-model integration system to comprehensively utilize the prediction results of multiple models.
[0169] Step S331, re-estimate the spatiotemporal correlation. First, calculate the Moran's I index to quantify the spatial autocorrelation. Then, construct the spatiotemporal variogram, use the anisotropic model to capture the spatial variation characteristics, and introduce the time component to characterize the temporal correlation attenuation.
[0170] Step S332: Adjust the dynamic weights. Define a dynamic weight function, taking into account the time variation of the initial weights and the observed residuals. Then, use the least squares method to estimate the weight function parameters so that the theoretical weights are as close as possible to the empirical weights.
[0171] Step S333: Update the graph structure. Use the Graphical Lasso algorithm to estimate the precision matrix, extract partial correlation coefficients from it and update the adjacency matrix of the graph. Then, apply the community detection algorithm (such as the Louvain method) to re-divide the subgraph structure to reflect the hierarchical characteristics of the system.
[0172] Step S334, capturing nonlinear associations. First, calculate mutual information and use the k-nearest neighbor method to estimate high-dimensional probability density. Then, estimate transfer entropy to capture the dynamic information flow between variables. Finally, construct a nonlinear association matrix, taking into account mutual information and transfer entropy.
[0173] Step S335: Establish a time-varying correlation model. First, perform sliding window analysis and repeat the correlation analysis in different time periods. Then, build a state space model to describe the dynamic evolution of the correlation matrix. Finally, use particle filtering to estimate the time-varying correlation and gradually optimize the estimation through prediction, update and resampling steps.
[0174] Step S336, optimize matrix consistency. Define a consistency measure to quantify the difference between the actual correlation matrix and the idealized matrix. Then, construct an optimization problem to improve the consistency of the matrix while keeping it close to the original estimate. Finally, use the projected gradient method to solve the optimization problem and iteratively update the correlation matrix.
[0175] Step S341, analyze parameter uncertainty. Calculate the posterior statistical characteristics of the parameters, including the mean, variance, and 95% confidence interval. Then, calculate the partial rank correlation coefficient (PRCC), rank the importance of the parameters based on the |PRCC| value, and identify the key parameters.
[0176] Step S342: Evaluate structural uncertainty. Apply the Bayesian Model Average (BMA) method to calculate the posterior probability of each candidate model. Then, construct a model ensemble prediction to comprehensively consider the results of multiple models. Finally, calculate the prediction variance to quantify the uncertainty caused by the model structure.
[0177] Step S343, analyzing input data uncertainty. First, perform perturbation analysis, add random perturbations to the input variables, and perform Monte Carlo simulation. Then, use the response surface method to build a polynomial regression model to describe the input-output relationship. Finally, perform variance decomposition, calculate the first-order and total sensitivity indexes, and quantify the relative importance of each input variable.
[0178] Step S344: Construct prediction intervals. First, sample from the posterior distribution of the parameters through parameter uncertainty propagation and run multiple model simulations. Then, perform residual resampling, assume that the residuals follow a specific distribution, and generate new predictions. Finally, calculate the predicted quantiles and construct a prediction interval that takes heteroscedasticity into account.
[0179] Step S345: Decompose and attribute uncertainty. Apply the ANOVA decomposition method to decompose the total variance into main effects and interaction effects. Then, use Shapley value analysis to quantify the contribution of each uncertainty source. Finally, perform information entropy decomposition to analyze the source and propagation of uncertainty from the perspective of information theory.
[0180] Step S346, evaluate the robustness and sensitivity of the model. First, perform a local sensitivity analysis and calculate the sensitivity coefficient. Then, apply the Morris screening method to generate the OAT trajectory and analyze the distribution characteristics of the basic effects. Finally, use the PAWN method to compare the conditional and unconditional cumulative distribution functions to quantify the impact of the parameters on the output.
[0181] Step S347: Visualize spatiotemporal uncertainty. Draw an error ellipse to show the spatial distribution of the prediction error. Then, create a collection summary map, using a color band to represent the uncertainty range. Finally, construct a probability density evolution map to show how the prediction distribution changes over time.
[0182] Step S348, analyze the propagation of uncertainty. Apply the Monte Carlo chain method to construct a causal chain and propagate uncertainty step by step. Then, perform moment independence analysis and use Taylor expansion to approximate high-order moments. Finally, apply fuzzy set theory to define the membership function of the input, use the extension principle to calculate the membership function of the output, and fully describe the propagation process of uncertainty in the model.
[0183] By retrieving the measured data of key nodes, correcting the spatiotemporal dynamic correlation matrix and the parameters of the hydrological and water ecological model, and conducting a comprehensive uncertainty analysis, the accuracy and reliability of the spatiotemporal dynamic simulation evaluation of water ecological products have been significantly improved. In terms of the acquisition and preprocessing of measured data at key nodes, a multi-source data acquisition system and real-time transmission technology are used to ensure the comprehensiveness and timeliness of the data. The modified Z-score method is used for outlier detection, which effectively identifies and processes abnormal data and improves the reliability of subsequent analysis. The spatial consistency test effectively marks suspicious data by calculating the correlation coefficient between nodes and neighboring nodes, which is particularly effective for discovering local anomalies (such as local rainstorms, sudden pollution incidents, etc.). For the problem of missing data, the cubic spline interpolation method is used to deal with short-term missing data, while for long-term missing data, the multivariate LSTM model is innovatively applied. This deep learning-based method can capture complex time dependencies and improve the estimation accuracy of long-term missing data, which is crucial to maintaining the temporal continuity of water ecological product evaluation. The introduction of Ensemble Kalman Filter (EnKF) for data assimilation effectively integrates observation data and model predictions, which not only improves the accuracy of state estimation, but also can update model parameters in real time, so that the model always maintains good adaptability to the actual situation. In terms of parameter calibration of hydrological and water ecological models, the Sobol method is first used to perform global parameter sensitivity analysis to identify the key parameters that have the greatest impact on model output, which reduces the number of parameters that need to be optimized and improves calibration efficiency. The NSGA-III algorithm is used for multi-objective parameter optimization, and a set of Pareto optimal solutions are obtained by considering the Nash-Sutcliffe efficiency coefficient, percentage deviation and Kling-Gupta efficiency. This multi-objective optimization method can find the best balance between different performance indicators and is particularly suitable for complex water ecosystem simulation. The DREAM algorithm is innovatively introduced for Bayesian parameter estimation to obtain the posterior distribution of parameters, which not only provides the optimal parameter value, but also quantifies the uncertainty of the parameter, laying the foundation for subsequent uncertainty analysis. By comparing the BIC values of different model structures, a multi-model integrated system is constructed, which effectively reduces the impact of model structure uncertainty on water ecological product evaluation. In terms of spatiotemporal dynamic correlation matrix correction, the Moran's I index and spatiotemporal variogram were recalculated, and the spatial autocorrelation structure was updated, which is crucial for accurately describing the spatial distribution pattern of water ecological elements. A dynamic weight function was defined, and the correlation strength was adjusted according to the node observation residuals, so that the correlation matrix could change dynamically over time and better reflect the non-stationary characteristics of the watershed system. The Graphical Lasso algorithm was used to estimate the precision matrix and update the adjacency relationship of the graph, which effectively captured the conditional independence between elements and provided a more accurate network structure. The mutual information and transfer entropy between nodes were calculated to capture nonlinear associations, which is particularly important for understanding complex ecosystem interactions.Sliding window analysis and state space modeling were applied to construct a time-varying correlation model, so that the correlation matrix can reflect short-term fluctuations and long-term trends, providing a powerful tool for evaluating the dynamic changes in the value of water ecological products. In terms of uncertainty analysis, the MCMC method was used to generate parameter posterior samples and comprehensively evaluate parameter uncertainty. The Bayesian model average (BMA) method was used to quantify the model structure uncertainty, effectively reducing the bias that may be caused by a single model. The uncertainty of input data was evaluated by perturbation analysis and response surface method, and the impact of observation errors on simulation results was fully considered. The residual resampling method was innovatively introduced to construct the prediction interval, taking into account heteroscedasticity and providing a more reliable uncertainty estimate. ANOVA decomposition and Shapley value analysis were used to decompose and attribute uncertainty, and the contribution of uncertainty from different sources was quantified, which is of great significance for identifying the key sources of uncertainty in the evaluation of water ecological products and guiding subsequent data collection and model improvement. In general, step S3 significantly improved the accuracy, reliability and adaptability of spatiotemporal dynamic simulation evaluation of water ecological products through a series of advanced data processing, model calibration and uncertainty analysis techniques.
[0184] According to one aspect of the present application, step S4 is specifically:
[0185] Step S41, use geographically weighted regression and Kriging interpolation methods to spatialize parameters, and expand the calibrated parameters to the entire basin; call the pre-built parallel computing framework and use MPI to perform domain decomposition; apply ensemble Kalman filtering to perform real-time data assimilation and update the model state; process upstream and downstream boundary conditions, use observation data to drive the upstream boundary, and apply hydraulic relations to define the downstream boundary;
[0186] Step S42: Evaluate water regulation services, calculate flood storage capacity and available water resources; quantify water purification services, estimate pollutant reduction and its economic value; evaluate biodiversity maintenance services, calculate habitat quality index and species richness; calculate carbon fixation and regulation services, estimate net ecosystem productivity and carbon storage changes; evaluate cultural services, including tourism and leisure value and aesthetic value; consider the interaction between services, use multi-criteria decision analysis methods to integrate various services, and obtain the overall ecosystem service value;
[0187] Step S43, construct a spatial scale sequence, use the area weighted average method for upper scale aggregation, and use a machine learning-based method for lower scale decomposition; construct a time scale sequence, use arithmetic mean and weighted average for time aggregation, and apply wavelet decomposition and EEMD methods for time decomposition; analyze scale effects, calculate scale sensitivity indexes, and estimate fractal dimensions; construct a multi-scale graph network, use a graph convolutional network to capture cross-scale features, and evaluate prediction uncertainties at different scales;
[0188] Step S44: Create time series animation and spatial heat map to show the spatiotemporal dynamic changes of the value of water ecological products; use error ellipse diagram and probability density evolution diagram to visualize uncertainty and form the value of water ecological products with uncertainty. Expand the calibrated model to the entire watershed scale to ensure the applicability and accuracy of the model in a wider range.
[0189] Steps S41 to S44 are specifically as follows:
[0190] Step S411, implement parameter spatialization. First, use the geographically weighted regression method to consider the impact of spatial location on parameters and build a parameter model of spatial variation. Then, apply the Kriging interpolation technique to estimate the parameter values of unknown locations based on the observed values of known points, while considering spatial autocorrelation.
[0191] Step S412: Execute scale conversion. For up-scale conversion, the area weighted average method is used to aggregate fine-scale data into coarse-scale data. For down-scale conversion, a machine learning model (such as random forest) is constructed in combination with high-resolution auxiliary data (such as digital elevation model, land use data) to decompose coarse-scale data into fine-scale data.
[0192] Step S413, build a parallel computing framework. First, use the METIS graph partitioning algorithm to divide the watershed into sub-regions to ensure load balancing and minimum communication overhead. Then, use MPI to implement inter-process communication, transfer boundary data and synchronize time steps. Finally, use OpenMP to implement fine-grained parallelism within a single computing node to accelerate unit-level computing.
[0193] Step S414: Implement model coupling and data assimilation. Use the OpenMI interface standard to define data exchange between components and implement coupling of sub-models such as hydrology, hydrodynamics, and water quality. Then, apply the ensemble Kalman filter method for data assimilation, and fuse model prediction and observation data through steps such as ensemble forecasting, observation forecasting, calculation of Kalman gain, and state updating.
[0194] Step S415, process boundary conditions. For upstream boundaries, the observation data is used first, and the unit line model is used when observation is missing. For downstream boundaries, the normal water depth relationship is used in non-tidal areas, and the harmonic analysis method is used in tidal areas.
[0195] Step S416, evaluate the model performance. Calculate multi-objective evaluation indicators, such as Nash-Sutcliffe efficiency coefficient and Kling-Gupta efficiency. Then, perform a spatial distribution consistency test, calculate Moran's I spatial autocorrelation index and SPAEF index, and comprehensively evaluate the spatiotemporal simulation capability of the model.
[0196] Step S42a, evaluate water regulation services. First, calculate the flood storage capacity, obtain the storage capacity by integrating the difference between inflow and outflow, and estimate the economic benefits. Then, evaluate the water resource supply service, calculate the available water volume and multiply it by the water price to obtain the economic value.
[0197] Step S42b, quantify the water purification service. Calculate the pollutant reduction amount, which is obtained by integrating the inflow and outflow pollutant concentrations with the flow rate. Then, use the substitution cost method to multiply the pollutant reduction amount by the equivalent treatment cost to estimate the economic value of the water purification service.
[0198] Step S42c, assess biodiversity maintenance services. First, calculate the habitat quality index, taking into account the impact of various threat factors. Then, apply the species-area relationship model to estimate species richness. Finally, use the habitat equivalence analysis method to assess the long-term value of biodiversity services.
[0199] Step S42d, quantify carbon fixation and regulation services. Calculate net ecosystem productivity, taking into account gross primary productivity, autotrophic respiration, and heterotrophic respiration. Then, estimate carbon stock change, taking into account net ecosystem productivity, harvest losses, and disturbance losses. Finally, multiply the carbon stock change by the carbon price or social carbon cost to obtain the economic value of carbon fixation services.
[0200] Step S42e, evaluate cultural services. Use the travel cost method to estimate the tourism and leisure value, taking into account the number of tourists, travel costs and willingness to pay. Apply the hedonic price method to evaluate the aesthetic value and analyze the impact of environmental characteristics on real estate prices. Evaluate the educational and scientific research value based on the number of published papers and research funds.
[0201] Step S42f, comprehensively evaluate the ecosystem service value. Construct a service interaction matrix to analyze the interactions between various services. Use the analytic hierarchy process to determine the weight of each service and calculate the comprehensive value. Finally, construct a Markov state transition matrix to predict the dynamic changes of future service values.
[0202] Step S431, perform spatial scale analysis. Construct a spatial scale sequence, such as 30m to 10km. For upper scale aggregation, use area weighted average or dominant type method. For lower scale decomposition, use machine learning methods (such as random forest or gradient boosting tree) for prediction based on high-resolution auxiliary data.
[0203] Step S432: Perform time scale analysis. Construct time scale series, such as day, month, season, and year. For time aggregation, use arithmetic mean or weighted mean method. For time decomposition, apply wavelet decomposition or ensemble empirical mode decomposition (EEMD) method to decompose long time series into different frequency components.
[0204] Step S433: Analyze the scale effect. Calculate the scale sensitivity index to assess the sensitivity of ecosystem services to scale changes. Estimate the fractal dimension to describe the spatial complexity of ecosystem services. Perform variogram analysis to fit a theoretical model to describe the characteristics of spatial correlation with scale changes.
[0205] Step S434: Study cross-scale feature transfer. Construct a multi-scale graph network, where nodes represent regions of different scales and edges represent relationships between scales. Use graph convolutional network modeling to capture cross-scale feature transfer. Introduce scale consistency constraints to ensure the coordination of prediction results at different scales.
[0206] Step S435: Perform multi-scale uncertainty analysis. Perform scale-dependent variance decomposition to separate the contribution of factors at different scales. Perform multi-scale sensitivity analysis to calculate the Sobol index at different scales and analyze the trend of sensitivity changing with scale. Evaluate the scale conversion error and calculate the root mean square error between the predicted value and the true value.
[0207] Step S441, realize multi-dimensional data visualization. Create a parallel coordinate graph, where each vertical axis represents a variable and each line represents a sample. Draw a radar chart, where multiple axes radiate outward from the center to represent different indicators. Generate a scatter matrix graph, and display the scatter plots and correlation coefficients between variables at the same time.
[0208] Step S442: Display spatiotemporal dynamic features. Create time series animations and use the gganimate package to create dynamic maps to display the temporal changes in ecosystem services. Draw spatial heat maps and use color gradients to represent the service values of different regions. Build a spatiotemporal cube and use the rgl package to create 3D visualizations to display changes in both spatial and temporal dimensions.
[0209] By deploying the hydrological and water ecological model with updated parameters, combined with the measured data of key nodes and the revised spatiotemporal dynamic correlation matrix, high-precision, multi-scale simulation and evaluation of the value of water ecological products are achieved. In terms of basin-scale deployment of the optimization model, the geographically weighted regression and Kriging interpolation methods are used to spatialize the parameters, which effectively solves the problem of generalizing point-scale parameters to the basin scale and ensures the applicability of the model in the entire basin. In particular, in the process of parameter space interpolation, auxiliary variables such as topography and land use are considered, which significantly improves the interpolation accuracy, which is crucial for accurately simulating the heterogeneous distribution of water ecological products in space. The constructed parallel computing framework combines MPI for domain decomposition and OpenMP to accelerate unit-level computing, which greatly improves the computational efficiency of the model. This high-performance computing method makes large-scale, high-resolution basin water ecological product simulation possible, providing technical support for long-term and dynamic evaluation. The realized model component coupling uses the OpenMI interface standard to define data exchange, which not only improves the modularity and scalability of the model, but also ensures the effective interaction between hydrological, ecological and socio-economic processes, which is crucial for a comprehensive assessment of the formation and value realization of water ecological products. The application of ensemble Kalman filtering for real-time data assimilation and continuous updating of model status enables the model to respond to dynamic changes in the watershed system in a timely manner, such as sudden rainfall events or human activity interference, and improves the timeliness and accuracy of water ecological product evaluation. In terms of water ecological product value simulation, a comprehensive evaluation framework has been developed, covering water regulation, water quality purification, biodiversity maintenance, carbon fixation and regulation, and cultural services. In the water regulation service evaluation, the hydrological model output and economic value evaluation method are innovatively combined, which not only calculates the flood storage capacity and available water resources, but also estimates its economic benefits, providing a direct economic basis for water resource management decisions. The water quality purification service evaluation adopts a method that combines pollutant reduction and substitution cost methods to convert the purification function of the ecosystem into a quantifiable economic value, which is of great significance for evaluating the environmental benefits of water ecosystems. In the biodiversity maintenance service evaluation, the habitat quality index, species richness model and habitat equivalence analysis method are innovatively combined to comprehensively evaluate the conservation value of the ecosystem. This method not only takes into account species diversity, but also incorporates habitat quality and ecosystem functions, providing a scientific basis for biodiversity conservation. The assessment of carbon fixation and regulation services realizes the economic value of ecosystem carbon sink function by calculating the net ecosystem productivity and carbon storage changes and combining carbon trading prices or social carbon costs, which is of great significance for evaluating the role of water ecosystems in climate change mitigation. The cultural service assessment innovatively combines the travel cost method, the characteristic price method and the bibliometric method to comprehensively evaluate the tourism, leisure, aesthetic and educational and scientific research value of water ecosystems. This multi-dimensional assessment method effectively captures the non-use value of water ecological products.In terms of multi-scale water ecological product evaluation, a spatial scale sequence from 30m to 10km was constructed, and the area weighted average method was innovatively applied for upper scale aggregation, and the machine learning-based method was used for lower scale decomposition. This multi-scale evaluation method not only solves the problem of information loss in the scale conversion process, but also provides evaluation results of corresponding scales according to different management and decision-making needs. In terms of time scale, a time scale sequence was constructed from day to year, and arithmetic mean and weighted average were used for time aggregation, and wavelet decomposition and EEMD methods were used for time decomposition. This method effectively captures the multi-time scale variation characteristics of the value of water ecological products and provides a scientific basis for short-term management and long-term planning. The scale effect analysis was innovatively introduced, the scale sensitivity index was calculated, and the fractal dimension was estimated, which deeply revealed the law of the change of water ecological product value with scale, which is of great significance for understanding and predicting ecosystem service functions at different scales. The construction of a multi-scale graph network and the use of a graph convolutional network to capture cross-scale features effectively solve the problem of information transmission between scales that is difficult to handle with traditional methods, and provides a new technical path for the evaluation of water ecological products in large-scale basins. In terms of result visualization and interpretation, multidimensional data visualization techniques, such as parallel coordinates, radar and scatter matrix plots, are used to intuitively display the multidimensional attributes of water ecological products, which helps decision makers fully understand the evaluation results. The time series animation and spatial heat map created dynamically display the spatiotemporal changes in the value of water ecological products, providing a powerful tool for identifying hot spots and critical periods. The use of error ellipse plots and probability density evolution plots to visualize uncertainty not only shows the credible interval of the evaluation results, but also reveals the spatiotemporal distribution characteristics of uncertainty, providing an important reference for risk assessment and management. The constructed interactive dashboard allows users to flexibly explore results at different scales and scenarios, improving the usability and decision-making support capabilities of the evaluation results.
[0210] According to one aspect of the present application, step S12 is specifically:
[0211] Step S121, divide the hydrological units, take the sub-basins as nodes at the macro scale, calculate the area, average elevation and proportion of main land use types of each sub-basin; use the hydrological observation data to identify key hydrological points at the meso scale, extract the longitude and latitude coordinates and related attributes such as flow and water quality parameters, divide the study area into regular grids at the micro scale, and assign unique identifiers and grid coordinates to each grid unit;
[0212] Step S122, using multi-source data, calculate the comprehensive attribute vector for each node; in terms of hydrological attributes, use long-term hydrological observation data to calculate the annual average flow, flow variation coefficient and peak flow; use groundwater monitoring data to extract groundwater level, aquifer thickness and permeability coefficient; in terms of ecological attributes, calculate the NDVI mean and standard deviation based on remote sensing data; use biodiversity survey data to calculate the Shannon index; combine remote sensing and field survey data to estimate biomass; in terms of socio-economic attributes, integrate population census and economic statistics to calculate population density, GDP and major industry types; use water use statistics and pollution source census data to calculate water consumption and major pollutant emissions;
[0213] Step S123, calculate the flow direction matrix based on the DEM data, determine the hydrological connection relationship between nodes, use the cumulative flow data to calculate the hydrological distance matrix, use the biodiversity data to construct the species similarity matrix, and combine the geographic distance to calculate the ecological distance matrix, integrate the transportation network data and the economic flow data, and construct the socio-economic connection matrix; integrate the hydrological distance matrix, the species similarity matrix and the socio-economic connection matrix, and use the multi-objective optimization method to determine the final edge weights to balance the hydrological connectivity, ecological similarity and socio-economic connection;
[0214] Step S124, calculate the centrality index of the node; comprehensively consider the role of the centrality index in hydrological regulation, ecological protection and socio-economic development, evaluate the hydrological and ecological importance of the node, and identify the key nodes based on the preset importance threshold; through the iterative merging process, merge the non-key nodes with the adjacent important nodes, and update the attributes and connection relationships of the merged nodes.
[0215] According to one aspect of the present application, step S13 is specifically:
[0216] Step S131: Based on the structure of the spatiotemporal dynamic graph of the watershed, five main modules are divided into hydrology, hydraulics, water quality, ecology and socio-economics, and standardized input and output interfaces are defined for each module;
[0217] Step S132: Use the node attribute data in the spatiotemporal dynamic graph of the watershed to provide an initial estimate of the parameters of each module; use the transfer learning method to construct the prior distribution of parameters based on the existing models of similar watersheds; apply the Bayesian inference framework and combine local observation data to update the posterior distribution of parameters; for parameters that are difficult to measure directly, use the DREAM algorithm to sample and estimate parameters; perform parameter sensitivity analysis and identify key parameters;
[0218] Step S133, construct a candidate model set, including sub-model combinations of different complexities, and construct multi-objective evaluation indicators, including simulation accuracy, computational efficiency, and model complexity; use a preconfigured algorithm to search for the optimal structure in the model space; for each candidate model, use cross-validation to evaluate its performance under different conditions; based on the trade-off between model performance and complexity, use the Akaike information criterion to select the optimal model structure;
[0219] Step S134: Use the MCMC method to generate parameter posterior samples, evaluate parameter uncertainty, and quantify model structural uncertainty through the Bayesian model averaging method; perform perturbation analysis on the input data to evaluate the impact of data uncertainty on model output; use the Sobol method to calculate the contribution of uncertainty from different sources; construct a prediction interval, considering the combined uncertainty of parameters, structure and input data; perform scenario analysis to evaluate the robustness of the model under different external conditions.
[0220] According to one aspect of the present application, step S14 is specifically:
[0221] Step S141, perform wavelet transform on the time series data of each node in the spatiotemporal dynamic diagram of the watershed to obtain wavelet coefficients at different time scales, calculate the cross-wavelet coherence between nodes at each time scale, and quantify the time-varying correlation between nodes; construct a multi-scale spatial weight matrix, calculate the global and local Moran's I index, and evaluate the spatial autocorrelation; combine the time and space dimensions to construct a spatiotemporal map, extract low-dimensional representations, and reveal the main spatiotemporal patterns of the system;
[0222] Step S142: Using long-term observation data, perform Granger causality test on each pair of nodes, identify potential causal relationships, calculate transfer entropy, capture nonlinear and indirect causal effects, and build an initial causal network based on the analysis results; apply the FCI algorithm and consider the influence of hidden variables to infer a more reliable causal structure; calculate the causal strength of each edge and use the threshold method to simplify the network structure; perform quantitative analysis on the retained causal relationships and estimate the average causal effect;
[0223] Step S143: Organize the data of the spatiotemporal dynamic graph of the watershed into a high-order tensor, use the Tucker decomposition method to reduce the dimension of the tensor, and obtain a core tensor and a factor matrix; based on the decomposition results, extract the main spatiotemporal patterns, apply sparse coding technology, and identify significant local features from low-dimensional representations; perform cluster analysis on the extracted patterns to reveal the main behavior types of the system, use the t-SNE algorithm to map the high-dimensional patterns to a two-dimensional plane, and intuitively display the similarities and evolution trends between the patterns;
[0224] Step S144, based on the topological structure of the spatiotemporal dynamic graph of the watershed, calculate the spatial proximity between nodes; use the hydrological model simulation results to quantify the hydrological connectivity between nodes; combine ecological survey data to evaluate the ecological similarity between nodes; integrate socioeconomic statistical data to calculate the strength of economic connections between nodes; use the fuzzy integral method to integrate the above types of association metrics to construct an initial association matrix; apply the spectral clustering method to identify the community structure and adjust the association strength accordingly; use the PageRank algorithm to calculate the node importance, weight the association matrix, and obtain the final comprehensive spatiotemporal association matrix.
[0225] According to one aspect of the present application, in step S42, the process of considering the interactions between services and using a multi-criteria decision analysis method to integrate various services to obtain the overall ecosystem service value is specifically as follows:
[0226] Step S421: Considering the interaction between services, a service interaction matrix is constructed, and the net effect is calculated. The net effect is the difference between the impact of the service on other services and the impact of other services.
[0227] Step S422: using the analytic hierarchy process to determine the weights of different services and obtain a weight vector, which contains the relative importance of each service; based on the weight vector, calculating the comprehensive value, that is, the weighted sum of the values of each service, which is the overall ecosystem service value;
[0228] Step S423: call a pre-built Markov state transition matrix for describing the probability of an ecosystem service transitioning from one state to another, and evaluate the evolution trend of the overall ecosystem service value in the future period.
[0229] According to another aspect of the present application, a spatiotemporal dynamic simulation and evaluation system for watershed water ecological products is also provided, comprising:
[0230] at least one processor; and,
[0231] a memory communicatively connected to at least one of the processors; wherein,
[0232] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products described in any of the above technical solutions.
[0233] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
Claims
1. Spatial-temporal dynamic simulation and evaluation method of watershed water ecological products, It is characterized in that The steps include: Step S1, determine the study area and obtain multi-source data, pre-process the multi-source data, form a pre-processed data set and fuse it, build a spatiotemporal dynamic map and a hydrological and water ecological simulation model for the study area based on the fused pre-processed data set, calculate the spatiotemporal correlation of the spatiotemporal dynamic map, and form a spatiotemporal correlation matrix; Step S2, read the spatiotemporal dynamic graph of the watershed, calculate the multidimensional importance index of each node in the graph, identify and extract key nodes at different spatial scales based on the node importance and graph structure characteristics, and consider the multi-scale characteristics of the watershed; Step S3, retrieve the measured data of key nodes, modify the spatiotemporal dynamic correlation matrix and the parameters of the hydrological and water ecological model through the measured data, and perform uncertainty analysis; Step S4: deploy the hydrological and water ecological model after parameter update, and simulate and evaluate the value of water ecological products by combining the measured data of key nodes and the spatiotemporal dynamic association matrix after parameter update, and output the simulation evaluation results; The step S2 is specifically as follows: Step S21, read the node attribute data in the spatiotemporal dynamic graph, calculate the importance indexes of the three dimensions of hydrology, ecology and socio-economy, use the hierarchical analysis method to determine the weight of each index, and comprehensively obtain the multidimensional importance index of each node; Step S22: sort all nodes in descending order based on the multidimensional importance index, select the first N nodes as the initial candidate set; calculate the graph topology features of the nodes; Construct node importance scores, combining multidimensional importance indices and graph topology features; Use a greedy algorithm to iteratively select the node with the highest score, and update the scores of the remaining nodes after each selection; calculate the spatial distribution index of the selected key nodes, and if it is lower than the threshold, adjust the node selection while maintaining the overall importance; N is a natural number; Step S23, constructing a spatial scale sequence; constructing a corresponding graph structure for each scale, and merging nodes in the same grid into super nodes; Calculate the features of the graph structure at each scale and the rate of change of features between adjacent scales; use the silhouette coefficient method to select the optimal scale; apply the key node identification algorithm on the optimal scale; recursively apply the identification algorithm to the key super nodes containing multiple original nodes; merge the key nodes of all levels to obtain the final multi-scale key node set; Step S24, dividing the total simulation time into a number of overlapping time windows, and calculating the time series characteristics of the nodes in each time window; Update the node importance score and reapply the key node identification algorithm; calculate the Jaccard similarity of the key node sets in adjacent time windows, and if the similarity is lower than the threshold, perform smoothing adjustment; analyze the frequency of nodes being selected as key nodes in all time windows, classify the nodes into stable key nodes, periodic key nodes, and instantaneous key nodes, and use the time series prediction model to predict the potential key nodes in the next time window; The step S3 is specifically as follows: Step S31, extract the measured data of key nodes, use the modified Z-score method to detect outliers, mark and process abnormal data; apply spatial consistency test, calculate the correlation coefficient between nodes and adjacent nodes, and mark suspicious data; for short-term missing data, use cubic spline interpolation method to supplement; for long-term missing, use multivariate LSTM model for estimation; use ensemble Kalman filter for data assimilation, fuse observation data and model prediction, and obtain the optimal estimate; Step S32: Use the Sobol method to perform global parameter sensitivity analysis, identify key parameters, construct a multi-objective function and apply the NSGA-III algorithm to perform multi-objective parameter optimization to obtain a set of Pareto optimal solutions; use the improved DREAM algorithm to perform Bayesian parameter estimation to obtain the posterior distribution of parameters; compare different model structures based on the BIC criterion and construct a multi-model integrated system; Step S33: based on the measured data of key nodes, recalculate the Moran's I index and spatiotemporal variogram, update the spatial autocorrelation structure, construct a dynamic weight function, and adjust the correlation strength according to the node observation residuals; Use the GraphicalLasso algorithm to estimate the precision matrix and update the adjacency relationship of the graph; Calculate mutual information and transfer entropy between nodes to capture nonlinear associations; Apply sliding window analysis and state space modeling to build a time-varying correlation model and obtain a revised spatiotemporal dynamic correlation matrix; Step S34, using the MCMC method to generate parameter posterior samples and calculate the posterior statistical characteristics of the parameters; applying the Bayesian model averaging method to quantify the model structure uncertainty; evaluating the input data uncertainty through perturbation analysis and response surface method; constructing the prediction interval, using the residual resampling method to consider heteroskedasticity; performing uncertainty decomposition and attribution, using ANOVA decomposition and Shapley value analysis to quantify the uncertainty contribution from different sources; The step S4 is specifically as follows: Step S41, use geographically weighted regression and Kriging interpolation methods to spatialize parameters, and expand the calibrated parameters to the entire basin; call the pre-built parallel computing framework and use MPI to perform domain decomposition; apply ensemble Kalman filtering to perform real-time data assimilation and update the model state; process upstream and downstream boundary conditions, use observation data to drive the upstream boundary, and apply hydraulic relations to define the downstream boundary; Step S42: Evaluate water regulation services, calculate flood storage capacity and available water resources; quantify water purification services, estimate pollutant reduction and its economic value; evaluate biodiversity maintenance services, calculate habitat quality index and species richness; calculate carbon fixation and regulation services, estimate net ecosystem productivity and carbon storage changes; evaluate cultural services, including tourism and leisure value and aesthetic value; consider the interaction between services, use multi-criteria decision analysis methods to integrate various services, and obtain the overall ecosystem service value; Step S43, construct a spatial scale sequence, use the area weighted average method for upper scale aggregation, and use a machine learning-based method for lower scale decomposition; construct a time scale sequence, use arithmetic mean and weighted average for time aggregation, and apply wavelet decomposition and EEMD methods for time decomposition; analyze scale effects, calculate scale sensitivity indexes, and estimate fractal dimensions; construct a multi-scale graph network, use a graph convolutional network to capture cross-scale features, and evaluate prediction uncertainties at different scales; Step S44, create time series animation and spatial heat map to show the spatiotemporal dynamic changes of the value of water ecological products; use error ellipse diagram and probability density evolution diagram to visualize uncertainty and form water ecological product value results with uncertainty.
2. According to the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products in claim 1, It is characterized in that The step S1 is specifically as follows: Step S11, determining a research area, acquiring and preprocessing multi-source data to obtain preprocessed multi-source data; fusing the preprocessed multi-source data through a data fusion method based on graph embedding; Step S12: constructing a spatiotemporal dynamic map of the watershed based on the preprocessed multi-source data; Step S13: establishing a hydrological and water ecological simulation model and obtaining simulation results for the constructed spatiotemporal dynamic diagram of the watershed; Step S14: Based on the simulation results and the spatiotemporal dynamic diagram of the watershed, perform spatiotemporal correlation analysis and construct a comprehensive spatiotemporal correlation matrix.
3. According to the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products in claim 2, It is characterized in that The step S12 is specifically as follows: Step S121, dividing the hydrological units, taking the sub-basins as nodes on a macro scale, and calculating the area, average elevation and proportion of main land use types of each sub-basin; Using hydrological observation data, key hydrological points are identified at the mesoscale, and their latitude and longitude coordinates and related attributes such as flow and water quality parameters are extracted. At the microscale, the study area is divided into a regular grid, and each grid cell is assigned a unique identifier and grid coordinates; Step S122, using multi-source data, calculate the comprehensive attribute vector for each node; in terms of hydrological attributes, use long-term hydrological observation data to calculate the annual average flow, flow variation coefficient and peak flow; use groundwater monitoring data to extract groundwater level, aquifer thickness and permeability coefficient; in terms of ecological attributes, calculate the NDVI mean and standard deviation based on remote sensing data; use biodiversity survey data to calculate the Shannon index; combine remote sensing and field survey data to estimate biomass; in terms of socio-economic attributes, integrate population census and economic statistics to calculate population density, GDP and major industry types; use water use statistics and pollution source census data to calculate water consumption and major pollutant emissions; Step S123, calculate the flow direction matrix based on the DEM data, determine the hydrological connection relationship between nodes, use the cumulative flow data to calculate the hydrological distance matrix, use the biodiversity data to construct the species similarity matrix, and combine the geographic distance to calculate the ecological distance matrix, integrate the transportation network data and the economic flow data, and construct the socio-economic connection matrix; integrate the hydrological distance matrix, the species similarity matrix and the socio-economic connection matrix, and use the multi-objective optimization method to determine the final edge weights to balance the hydrological connectivity, ecological similarity and socio-economic connection; Step S124, calculating the centrality index of the node; comprehensively considering the role of the centrality index in hydrological regulation, ecological protection and socio-economic development, evaluating the hydrological and ecological importance of the node, and identifying the key nodes based on a preset importance threshold; Through the iterative merging process, non-critical nodes are merged with adjacent important nodes, and the attributes and connection relationships of the merged nodes are updated.
4. According to the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products in claim 2, It is characterized in that The step S13 is specifically as follows: Step S131: Based on the structure of the spatiotemporal dynamic graph of the watershed, five main modules are divided into hydrology, hydraulics, water quality, ecology and socio-economics, and standardized input and output interfaces are defined for each module; Step S132: Use the node attribute data in the spatiotemporal dynamic graph of the watershed to provide an initial estimate of the parameters of each module; use the transfer learning method to construct the prior distribution of parameters based on the existing models of similar watersheds; apply the Bayesian inference framework and combine local observation data to update the posterior distribution of parameters; for parameters that are difficult to measure directly, use the DREAM algorithm to sample and estimate parameters; perform parameter sensitivity analysis and identify key parameters; Step S133, construct a candidate model set, including sub-model combinations of different complexities, and construct multi-objective evaluation indicators, including simulation accuracy, computational efficiency, and model complexity; use a preconfigured algorithm to search for the optimal structure in the model space; for each candidate model, use cross-validation to evaluate its performance under different conditions; based on the trade-off between model performance and complexity, use the Akaike information criterion to select the optimal model structure; Step S134: Generate parameter posterior samples using the MCMC method to evaluate parameter uncertainty, and quantify model structure uncertainty using the Bayesian model averaging method; perform perturbation analysis on input data to evaluate the impact of data uncertainty on model output; The Sobol method is used to calculate the contribution of uncertainty from different sources; Construct prediction intervals, taking into account the combined uncertainties of parameters, structure and input data; conduct scenario analysis to assess the robustness of the model under different external conditions.
5. According to the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products in claim 2, It is characterized in that The step S14 is specifically as follows: Step S141, perform wavelet transform on the time series data of each node in the spatiotemporal dynamic diagram of the watershed to obtain wavelet coefficients at different time scales, calculate the cross-wavelet coherence between nodes at each time scale, and quantify the time-varying correlation between nodes; construct a multi-scale spatial weight matrix, calculate the global and local Moran's I index, and evaluate the spatial autocorrelation; combine the time and space dimensions to construct a spatiotemporal map, extract low-dimensional representations, and reveal the main spatiotemporal patterns of the system; Step S142: Using long-term observation data, perform Granger causality test on each pair of nodes, identify potential causal relationships, calculate transfer entropy, capture nonlinear and indirect causal effects, and build an initial causal network based on the analysis results; apply the FCI algorithm and consider the influence of hidden variables to infer a more reliable causal structure; calculate the causal strength of each edge and use the threshold method to simplify the network structure; perform quantitative analysis on the retained causal relationships and estimate the average causal effect; Step S143: Organize the data of the watershed spatiotemporal dynamic graph into a high-order tensor, use the Tucker decomposition method to reduce the dimension of the tensor, and obtain a core tensor and a factor matrix; based on the decomposition result, extract the main spatiotemporal patterns, and apply sparse coding technology to identify significant local features from the low-dimensional representation; Perform cluster analysis on the extracted patterns to reveal the main behavior types of the system. Use the t-SNE algorithm to map the high-dimensional patterns to a two-dimensional plane to intuitively display the similarities and evolution trends between the patterns. Step S144, calculating the spatial proximity between nodes based on the topological structure of the watershed spatiotemporal dynamic graph; Using hydrological model simulation results, quantify the hydrological connectivity between nodes; Combine ecological survey data to assess ecological similarity between nodes; integrate socioeconomic statistics to calculate the strength of economic connections between nodes; The fuzzy integral method is used to integrate the above types of association metrics to construct the initial association matrix; the spectral clustering method is applied to identify the community structure and adjust the association strength accordingly; the PageRank algorithm is used to calculate the node importance and weight the association matrix to obtain the final comprehensive spatiotemporal association matrix.
6. According to the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products in claim 2, It is characterized in that In step S42, the process of considering the interaction between services and using a multi-criteria decision analysis method to integrate various services to obtain the overall ecosystem service value is as follows: Step S421: Considering the interaction between services, a service interaction matrix is constructed, and the net effect is calculated. The net effect is the difference between the impact of the service on other services and the impact of other services. Step S422: using the analytic hierarchy process to determine the weights of different services and obtain a weight vector, which contains the relative importance of each service; based on the weight vector, calculating the comprehensive value, that is, the weighted sum of the values of each service, which is the overall ecosystem service value; Step S423: call a pre-built Markov state transition matrix for describing the probability of an ecosystem service transitioning from one state to another, and evaluate the evolution trend of the overall ecosystem service value in the future period.
7. A spatiotemporal dynamic simulation and evaluation system for watershed water ecological products. It is characterized in that include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the spatiotemporal dynamic simulation and evaluation method of watershed water ecological products as described in any one of claims 1 to 6.
Citation Information
Cited By
Data fusion method and system based on reservoir multi-source data
CN119475254A
Data fusion method and system based on multi-source data of reservoir
CN119475254B
