Multi-dimensional diagnosis method for river ecological degradation degree

Abstract

The invention discloses a multi-dimensional diagnosis method for river ecological degradation degree, and particularly relates to the technical field of river ecological degradation analysis. Obtaining multiple types of ecological parameters of the target river reach, and constructing a physical and chemical coupling parameter set, a biocenosis response factor set, a structure instability index set and an ecological function coupling parameter set; performing wavelet transform and principal component merging on the parameter set to form a multi-dimensional ecological feature tensor; extracting a main disturbance feature subspace by adopting high-order singular value decomposition to obtain a disturbance feature vector set; constructing a nonlinear degradation function based on an ecological stress response mechanism, outputting an ecological degradation risk value, identifying an ecological degradation weak region coordinate set, and generating a spatial degradation trend chart and an intervention priority suggestion path; the river ecological degradation degree can be comprehensively and accurately revealed, the scientificity and interpretability of diagnosis and the pertinence of treatment are improved, and the method has remarkable application value.

Description

Technical Field

[0001] This invention relates to the field of river ecological degradation analysis technology, specifically to a multidimensional diagnostic method for the degree of river ecological degradation. Background Technology

[0002] As human activities increasingly disrupt river ecosystems, river ecological degradation is becoming increasingly serious, manifesting in various aspects such as water pollution, declining biodiversity, loss of riparian function, and disappearance of benthic organisms. Current methods for diagnosing river ecosystems mostly rely on single indicators, such as water quality parameters (nitrogen and phosphorus concentrations, dissolved oxygen, etc.) and biological indicators (fish or benthic animal diversity, etc.), which are insufficient to comprehensively reflect the true degree of ecosystem degradation.

[0003] Especially in some micro-river network areas (such as urban low-level tributaries, irrigation canals, and sand mining impact zones), the water flow is slow, pollution accumulation is strong, and the biological system is extremely unstable. Traditional single-indicator diagnostic methods cannot effectively identify subtle changes in the degree of ecological degradation, resulting in a lack of targeted ecological governance measures and even the phenomenon of "misjudging as good or bad".

[0004] Furthermore, existing methods generally lack coupling modeling mechanisms between indicators of different dimensions (physicochemical, biological, structural, and functional), failing to achieve "quantitative diagnosis" and "grading" of ecological degradation. Therefore, this invention proposes a system diagnostic method that integrates multi-dimensional ecological characteristic indicators, capable of identifying degradation trends and weak links in ecosystems at the microscale, thereby providing precise evidence for ecological restoration. Summary of the Invention

[0005] The purpose of this invention is to provide a multidimensional diagnostic method for the degree of river ecological degradation, in order to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multidimensional diagnostic method for the degree of river ecological degradation, comprising: Obtain the basic ecological parameter set of the target river section, including the multiphase physicochemical coupling parameter set P′, the variable biological community response factor set B′, the multiscale structural instability index set H′, and the ecological function coupling parameter set F′; P′, B′, H′ and F′ are respectively subjected to wavelet transform and principal component merging to construct the multidimensional ecological feature tensor T of the target river segment; Using the multidimensional tensor decomposition algorithm, the dominant perturbation feature subspace C in tensor T is extracted to obtain the set of dominant perturbation feature vectors. n is the total number of principal perturbation eigenvectors; Construct a nonlinear degradation function based on the ecological stress response mechanism and output the ecological degradation risk value of the target river section; Based on the abrupt change points of the ecological degradation risk value, the ecological degradation level is divided into 5 levels using the abrupt change segment threshold method: mild, mild-moderate, moderate, severe, and extremely severe. The river sections with severe and extremely severe degradation levels were subjected to reverse feature projection, and hot spot clustering was performed based on the distribution of main disturbance features to obtain the coordinate set of ecologically weak areas. Output a multi-dimensional ecological degradation diagnostic report, including a characteristic source tracing matrix, a spatial degradation trend map, and recommended intervention priority paths.

[0007] Preferably, the set of basic ecological parameters includes: a multiphase physicochemical coupling parameter set P′, including water temperature-dissolved oxygen co-response curves at multiple points within the same period, dynamic ratios of nitrogen and phosphorus, and turbidity-particulate matter correlation factors; a set of variable biological community response factors B′, including watershed-specific taxa response indices, the ratio of pollution-sensitive to pollution-tolerant species, and microbial community Beta diversity indicators; a set of multiscale structural instability indicators H′, including riverbank damage and fracture density, tributary inflow-confluence disturbance index, and cross-sectional hydrodynamic asymmetry coefficient; and a set of ecological function coupling parameters F′, including primary productivity variation rate, bioenergy conversion efficiency, and nitrogen cycle coupling index.

[0008] Preferably, the multidimensional ecological feature tensor T of the target river section includes: The wavelet feature vector group is input into the principal component analysis model, and the top k principal component vectors with a cumulative contribution rate greater than 90% are retained; the loading coefficients of the principal component vectors are used to quantify the contribution of each original parameter in the principal component; the principal component vectors are normalized to obtain the ecological disturbance principal vector set under a unified scale. The merged principal vector set is represented by a matrix to correspond to the four parameter sets, and four principal component sub-matrices MP, MB, MH, and MF are constructed with dimensions m×k, where m is the number of monitored samples and k is the number of principal components. The principal component submatrices MP, MB, MH, and MF are concatenated into tensors with dimension alignment; a third-order tensor T∈ is constructed using tensor stacking operations. The third dimension is used to distinguish the principal component expressions of the four types of parameter sets.

[0009] Preferably, the extraction of the dominant perturbation feature subspace C in tensor T includes: Perform high-order singular value decomposition (HOSVD) on the constructed multidimensional ecological feature tensor T; The tensor T is decomposed into a core tensor G and three orthogonal matrices U1, U2, and U3, which represent the orthogonal feature spaces of the sample dimension, principal component dimension, and feature class dimension, respectively. The core tensor G is used to characterize the joint contribution strength of each perturbation factor, and the column vectors in the orthogonal matrix U2 are the main perturbation eigenvector set C; The total number of vectors n in subspace C is determined based on the principle that the cumulative contribution rate reaches 95%.

[0010] Preferably, the HOSVD decomposition process employs a truncation decomposition strategy, including: Calculate the nth singular values ​​of tensor T in each dimension and sort them in descending order; Select the feature vectors corresponding to the first r singular values ​​of each dimension to construct a subspace, where r is the minimum dimension required to satisfy the cumulative contribution rate not being lower than a set threshold. The core tensor G is compressed and preserved, retaining only the tensor block corresponding to the r-dimensional dimension, and removing perturbation factors with low contribution.

[0011] Preferably, the construction of the nonlinear degradation function based on the ecological stress response mechanism includes: Select all or part of the vectors in the main perturbation feature subspace C as input terms of independent variables; Construct an ecological stress response index set R, including the abundance of mutant groups, functional loss factors, and the frequency of redox potential extreme values; A nonlinear regression modeling strategy is adopted to fit the mapping relationship between feature vectors and stress response indicators; The fitting function, R = f(C), is used to predict the ecological degradation risk value Rv of the target river section.

[0012] Preferably, the classification based on abrupt changes in ecological degradation risk values ​​includes: The ecological degradation risk value sequence of the target river section is sorted according to time or spatial dimensions; The rate of change of adjacent risk values ​​is calculated using the sliding window first-order difference method. When the rate of change exceeds the set threshold λ, the location is determined to be a risk value mutation point; Using all mutation points as interval dividing nodes, the risk value interval is divided into several segments; Arrange the segments after dividing the risk value range in numerical order; Calculate the mean and boundary points of each segment interval; Using the boundary values ​​of the segmented intervals as the threshold for classifying the severity, five severity levels are obtained: mild, mild-moderate, moderate, severe, and extremely severe.

[0013] Preferably, the reverse feature projection includes: A mapping relationship is established between the risk values ​​corresponding to the river sections identified as severe and extremely severe and their main disturbance feature vectors; Using the core tensor G and orthogonal matrix The reverse product method maps the risk value back to the main perturbation feature subspace; The characteristic distribution vector of the river segment on the multidimensional perturbation factor is obtained.

[0014] Preferably, the hotspot clustering identification based on the main perturbation feature distribution includes: Perform standardization on the feature distribution vector set obtained by back projection; The density clustering algorithm DBSCAN is used to identify high-density feature regions. The neighborhood radius ε is set to 1.5 times the mean square error of the sample features, and the minimum number of samples MinPts is 5. The distribution points identified as core samples and their neighborhoods are marked as potential hot spot regions. Spatial localization methods for potential hot spot regions include: extracting the coordinates of the monitoring section corresponding to each feature distribution point; By performing minimum bounding rectangle fitting on the coordinate set within the same cluster, the spatial boundary of the degenerate region can be obtained. The coordinates of the center point of the output boundary are used as the coordinate set of ecologically degraded and vulnerable areas.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention utilizes innovative techniques such as multidimensional ecological feature tensor modeling, principal perturbation feature extraction, nonlinear degradation function construction, and abrupt threshold classification to achieve quantitative, systematic, and spatial diagnosis of river ecological degradation. Compared to traditional evaluation methods that rely on single water quality or biological indicators, this invention can capture the coupling effects between multi-source heterogeneous parameters, revealing the true degradation patterns of ecosystems across multiple dimensions (physicochemical, biological, structural, and functional), thereby improving the scientific rigor and accuracy of the diagnosis.

[0016] 2. This invention accurately locates severely and extremely severely degraded weak areas through reverse feature projection and hotspot clustering identification. Combined with a feature source tracing matrix and a spatial degradation trend map, it outputs priority intervention paths, possessing causal interpretability and operability for governance. This method not only improves the sensitivity and robustness of degradation identification but also provides quantitative evidence for the optimal allocation of resources for watershed ecological restoration, pollution control, and governance. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, please refer to Figure 1 As shown in this embodiment, a multidimensional diagnostic method for the degree of river ecological degradation includes: Obtain the basic ecological parameter set of the target river section, including the multiphase physicochemical coupling parameter set P′, the variable biological community response factor set B′, the multiscale structural instability index set H′, and the ecological function coupling parameter set F′; P′, B′, H′ and F′ are respectively subjected to wavelet transform and principal component merging to construct the multidimensional ecological feature tensor T of the target river segment; Using the multidimensional tensor decomposition algorithm, the dominant perturbation feature subspace C in tensor T is extracted to obtain the set of dominant perturbation feature vectors. n is the total number of principal perturbation eigenvectors; Construct a nonlinear degradation function based on the ecological stress response mechanism and output the ecological degradation risk value of the target river section; Based on the abrupt change points of the ecological degradation risk value, the ecological degradation level is divided into 5 levels using the abrupt change segment threshold method: mild, mild-moderate, moderate, severe, and extremely severe. The river sections with severe and extremely severe degradation levels were subjected to reverse feature projection, and hot spot clustering was performed based on the distribution of main disturbance features to obtain the coordinate set of ecologically weak areas. Output a multi-dimensional ecological degradation diagnostic report, including a characteristic source tracing matrix, a spatial degradation trend map, and recommended intervention priority paths.

[0021] To achieve a multidimensional diagnosis of river ecological degradation, this invention first establishes a set of basic ecological parameters for the target river section, comprehensively reflecting its ecological characteristics across multiple dimensions, including physicochemical state, biological community, structural stability, and ecological function. Unlike traditional methods that use only a few water quality or biological indicators, the parameter set proposed in this invention exhibits multiphase heterogeneity, multidimensional coupling, and dynamic responsiveness. Specifically, it includes the following four types of sub-parameter sets: a multiphase physicochemical coupling parameter set P′, a set of variable biological community response factors B′, a set of multiscale structural instability indicators H′, and a set of ecological function coupling parameters F′.

[0022] Construction of the multiphase physicochemical coupling parameter set P′: This set is used to describe the synergistic or coupling relationship between various physicochemical factors in rivers under different spatiotemporal distributions, reflecting the physicochemical response mechanism of the water environment system under the background of ecological disturbance.

[0023] The construction steps are as follows: Using hourly timescales, at least five online water quality monitoring points are set up in the target river section to collect basic parameters such as water temperature, dissolved oxygen, total nitrogen, total phosphorus, turbidity, and suspended particulate matter concentration, forming a spatiotemporal multi-point dataset.

[0024] Construction of the water temperature-dissolved oxygen co-response curve: The correlation between water temperature and dissolved oxygen within the same time period was calculated using the Pearson correlation coefficient, and a moving window regression model was performed to generate the response curve. The intensity of the co-response was characterized by the maximum correlation coefficient within the moving window.

[0025] Dynamic nitrogen-phosphorus ratio: The mass ratio of total nitrogen to total phosphorus is calculated for each monitoring point in hourly data to form a time series. Wavelet transform is then applied to this series to extract its periodic changes and abrupt change features, which are used as the dynamic nitrogen-phosphorus ratio factor.

[0026] Turbidity-particulate matter correlation factor: Principal component regression (PCR) was performed on turbidity and suspended particulate matter concentration. The correlation index was constructed using the principal component contribution rate and the least squares value of the residuals, and it is expressed as "physicochemical particle coupling strength".

[0027] Finally, the above indicators were uniformly normalized to the [0,1] interval and constructed into a vector form: P′= {WDO_corr, N_P_ratio_var, Turb_SPM_index}, which correspond to the water temperature-dissolved oxygen synergy, nitrogen-phosphorus dynamic ratio, and turbidity-particulate matter correlation factor, respectively.

[0028] Construction of a set of variable biological community response factors B′, which is used to describe the composition and structure of biological communities in aquatic bodies, especially microorganisms, benthic animals, etc., and their sensitivity and response intensity to ecological disturbances.

[0029] The construction steps are as follows: No fewer than three biological monitoring sections were set up in the target river section to collect benthic animal and microbial samples. The samples were then classified and their abundance was statistically analyzed using metagenomic sequencing and microscopic examination.

[0030] Construction of the Watershed-Specific Taxonomic Response Index (LTRI): The response index is calculated by comparing the endemic species in the river section with the ecological tolerance coefficients in the regional reference database and combining them with the abundance change rate. It is defined as the abundance of the endemic taxonomic group multiplied by the ecological tolerance weighting factor.

[0031] The ratio of pollution-sensitive to pollution-tolerant species: EPT species (mayflies, caddisflies, and tauroptera) are considered pollution-sensitive, while annelids and worms are considered pollution-tolerant. The proportion of each species is calculated, and the ratio of the two categories is used as the factor value.

[0032] Microbial community Beta diversity index calculation: The Bray-Curtis distance is used to measure the differences in microbial community structure among different sampling points. The larger the Beta diversity value, the stronger the community instability.

[0033] The final parameter set is: B′={LTRI_index, Sens_Tol_ratio, Beta_diversity}.

[0034] Construction of a multi-scale structural instability index set H′: This set is used to reflect the stability of the target river segment in terms of morphological structure, especially the degree of disturbance in the riverbank, cross section and tributary confluence area.

[0035] The specific construction method is as follows: Riverbank damage and fracture density calculation: UAV imagery and image recognition algorithms (such as Mask R-CNN) are used to identify damaged areas of the shoreline, and the number of fractures per unit length and the total length are calculated as density indicators.

[0036] Tributary inflow-confluence disturbance index: Combining DEM hydrological analysis and flow velocity monitoring, the flow velocity abrupt change rate and water quality gradient change rate at the tributary inflow-confluence are calculated, and the two are weighted into a disturbance index.

[0037] Cross-sectional hydrodynamic asymmetry coefficient: The asymmetry coefficient is constructed by statistically analyzing the difference in flow velocity and the flow direction angle between the left and right banks of the same cross-section; the higher the value, the stronger the hydraulic disturbance instability of the cross-section.

[0038] The final parameter set is: H′ = {Bank_fracture_density, Inlet_disturbance_index,Flow_asymmetry_coefficient}.

[0039] Construction of the ecological function coupling parameter set F′: This set reflects the operational efficiency and coupling state of the river system at the ecological function level, such as energy conversion and material cycling.

[0040] The construction steps are as follows: Primary productivity variation rate calculation: The total daily primary productivity is calculated using photosynthesis monitoring data, and the daily average deviation is analyzed to extract the diurnal volatility index.

[0041] Bioenergy conversion efficiency calculation: The ratio of primary productivity to benthic animal biomass growth rate per unit time is calculated to form an energy conversion efficiency index, which is used to characterize energy flow efficiency.

[0042] Nitrogen cycle coupling index calculation: A coupling model is established based on the denitrification rate, ammonification rate, and total nitrogen loss rate. The goodness of fit between the actual observed rate and the theoretical ideal rate is calculated to characterize the coordination of the nitrogen cycle process.

[0043] The final parameter set is: F′={NPP_variation_rate, Energy_transfer_efficiency,Nitrogen_cycle_coupling_index}.

[0044] The above four types of sub-parameter sets form the following structures respectively: P′ (physicochemical parameters): a 3-dimensional vector; B′ (Biological Response): A 3-dimensional vector; H′ (structural stability): a 3-dimensional vector; F′ (ecological function): a 3-dimensional vector; Each set of vectors is uniformly Min-Max normalized to the interval [0,1], and then concatenated into a complete multidimensional ecological feature tensor for subsequent tensor modeling and degradation feature extraction.

[0045] Wavelet transform is a nonlinear signal processing technique that combines time-domain and frequency-domain analysis capabilities, and is particularly suitable for extracting abrupt changes and local disturbances in non-stationary signals. Considering that ecological parameters (such as water temperature, dissolved oxygen, and nutrients) often exhibit both seasonal fluctuations and sudden pollution events, wavelet decomposition can effectively capture their multi-scale disturbance information.

[0046] This invention selects the Daubechies wavelet function (denoted as db4) as the basic wavelet because it has good compact support and orthogonality, making it suitable for processing signal characteristics containing noise, fluctuations, and abrupt changes in ecological data. The length parameter of the Daubechies wavelet is set to order 4 to ensure a balance between smoothness and accuracy.

[0047] For each set of parameters (P′, B′, H′, and F′), a three-level Discrete Wavelet Transform (DWT) is performed. This decomposition divides each original sequence signal into one approximate component and three detail components, representing the low-frequency trend and high-frequency disturbance, respectively.

[0048] The following feature extraction process is performed on each component after wavelet decomposition: Approximate component (3rd layer): Extracts the number of local extrema and the trend slope of the sequence at the current scale to represent long-term stability; Detail components (layers 1 to 3): their standard deviation and energy density are calculated as indicators of the intensity of the mutation response; Each parameter is ultimately transformed into a set of 5-dimensional wavelet feature vectors, containing one trend indicator and four perturbation indicators.

[0049] For example, the original water temperature sequence can be converted into a wavelet feature vector W_temp = {local_max_count,std_d1, energy_d1, std_d2, energy_d2}, where d1 and d2 represent the detail components of the first and second layers, respectively.

[0050] The wavelet processing described above is performed on all parameters with temporal structure in P′, B′, H′, and F′ to obtain wavelet feature vector groups, denoted as WP′, WB′, WH′, and WF′, respectively. Their dimensions depend on the number of parameters and the number of decomposition levels. Each group of vectors is then normalized to unify their dimensions.

[0051] While multidimensional wavelet features are highly expressive, they suffer from redundancy, high collinearity, and difficulty in visualization. Therefore, Principal Component Analysis (PCA) is employed to compress the dimensions of each parameter set and fuse features to extract the most representative perturbation directions.

[0052] For the wavelet feature matrix of each parameter set, perform the following steps sequentially: Standardization: Subtract the mean from each column of data and divide by the standard deviation to ensure balanced input of all features; Covariance matrix calculation: Calculate the covariance matrix under the sample dimension to reflect the correlation between features; Eigenvalue and eigenvector decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain the principal component directions and their contributions; Principal component selection: retain the top k principal components so that their cumulative contribution rate reaches more than 90%, ensuring sufficient information retention and effective dimensionality reduction.

[0053] For example, if WP′ contains 10-dimensional wavelet features, the first 3 principal components can be retained after PCA. Its principal component matrix is: MP = [PC1_P, PC2_P, PC3_P], with a dimension of n×3, where n is the number of samples.

[0054] Following the above method, PCA processing was performed on WP′, WB′, WH′, and WF′ respectively to obtain four principal component matrices: MP: Principal component matrix of physicochemical properties; MB: Principal Component Matrix of Biology; MH: Principal Component Matrix of Structure; MF: Functional Principal Component Matrix; Each matrix has a dimension of m × k, where m is the number of samples and k is the number of principal components (e.g., uniformly 3 dimensions).

[0055] To integrate the principal component information of four types of ecological characteristics, a third-order tensor T is constructed, with the following structure: First dimension (m): Number of time or space nodes in the sample; Second dimension (k): Number of principal components; The third dimension (4) corresponds to four sets of parameters (physicochemical, biological, structural, and functional).

[0056] That is, the dimension of tensor T is: T∈ .

[0057] MP, MB, MH, and MF are merged by concatenating (stacking) the third dimension to form a tensor structure; all elements are stored in a standard numerical tensor data format after being uniformly normalized; tensor T provides the input basis for subsequent perturbation feature extraction, higher-order singular value decomposition (HOSVD), and ecological risk prediction.

[0058] Tensor T is stored in NumPy tensor format or TensorFlow tensor structure, has high-performance computing support, and can be directly used in the Python environment for tensor decomposition, cluster recognition and visualization analysis.

[0059] HOSVD is a generalization of the traditional singular value decomposition in high-dimensional tensor spaces. It is used to decompose a high-order tensor into: a core tensor G (representing the coupling strength between features); and multiple orthogonal matrices U1, U2, and U3, which represent the eigenvectors of tensor T in each dimension.

[0060] For the third-order tensor T in this invention, its HOSVD form can be expressed as: Where: G is the core tensor; These represent product operations in the 1st, 2nd, and 3rd dimensions, respectively. These are orthogonal matrices in the corresponding dimensions; The truncation rank selected for each dimension is used to control the amount of dimensionality reduction retained.

[0061] This invention employs the following steps to perform high-order singular value decomposition on tensor T and extract the perturbation principal eigenvectors: Expand the tensor into two-dimensional matrices along the three dimensions, denoted as T(1), T(2), and T(3): : Indicates the expansion of sample dimensions; : Indicates the expansion of the principal component dimensions; : Indicates the expansion of parameter category dimensions.

[0062] Performing standard singular value decomposition on each expanded matrix yields: ; ; ; in, It is a left singular matrix (orthogonal matrix) for each dimension, representing the direction of the principal axis in each dimension.

[0063] For each S matrix, the cumulative contribution rate is calculated for all singular values. A contribution rate threshold of 95% is set, and the corresponding rank is selected. , where represents the minimum number of dimensions that are sufficient to retain 95% of the information in each dimension.

[0064] For example: if the sample dimension n = 200, then it is possible to select = 10; If the principal component dimension k = 3, then = 2; If the parameter category dimension = 4, then = 2.

[0065] Calculate the core tensor G based on the selected orthogonal basis vectors: This core tensor describes the coupling strength and dominant factors between different perturbation directions.

[0066] From the above process Extract the orthogonal eigenvectors along the principal component dimension, denoted as the principal perturbation eigenspace C. n is the total number of principal perturbation eigenvectors. , which is the total number of main perturbation vectors retained; each vector , representing a higher-order perturbation direction, which integrates the response trend of multiple ecological parameter dimensions.

[0067] These vectors can be used as input variables in subsequent ecological degradation risk assessment models, and can also be used for perturbation mechanism analysis and cluster typing.

[0068] Ecological stress response refers to the phenomenon where certain organisms or functional parameters in an ecosystem exhibit significant abnormalities after being affected by external disturbances (such as pollution loads, structural damage, biological invasions, etc.). This type of response generally possesses the following characteristics: It is sensitive to specific disturbances and has threshold properties; Most of these are non-linear relationships, such as a sharp drop in biomass after the ecological balance is disrupted. It has a time delay or cumulative effect.

[0069] This invention combines the principles of ecology and ecotoxicology, and sets the following three types of representative ecological stress response indicators to construct a set of monitoring variables for the objective function, as follows: Abundance Shock Ratio (ASR): This index represents the abnormal increase or decrease of a certain key ecological group (such as EPT, indicator species, or dominant species) in a short period of time in a target river section.

[0070] The calculation method is as follows: ASR = current monitoring value / historical monthly average value. If the ASR is greater than 2.0 or less than 0.5, it is considered that a sudden change in ecological abundance has occurred, constituting a positive stress response.

[0071] Functional Deficit Index (FDI): Indicates whether key parameters related to core ecosystem functions (such as nitrogen cycle and primary productivity) are below a warning threshold.

[0072] For example, when the "primary productivity variation rate" continuously exceeds its ecological tolerance (set to ±30%) and the bioenergy conversion efficiency decreases by more than 40%, the FDI is recorded as 1, indicating the presence of functional deficit stress; otherwise, it is 0.

[0073] Redox Stress Frequency (RSF): This refers to the frequency at which the redox potential (ORP) of water exceeds ±250 millivolts within a certain time window (e.g., 7 days). High-frequency, large fluctuations usually indicate that the system is in an unstable or oxygen-depleted state.

[0074] The calculation method is as follows: the number of times the upper and lower thresholds (±250 mV) are exceeded within the statistical time window. If the frequency is ≥ 5 times / 7 days, it is considered a high-risk stress. The above three indicators constitute the ecological stress response indicator set R = {ASR, FDI, RSF}, which serves as the target variable set for the degradation risk model.

[0075] Based on the obtained main perturbation feature subspace C, where each perturbation vector This represents the combination of characteristics of an ecosystem under a specific disturbance mechanism. (Setting) , where each vector , which is a combination of k-dimensional features.

[0076] Select the top m vectors with the highest contribution rates as modeling inputs, preferably m = 5 to 10, to avoid overfitting caused by high-dimensional inputs.

[0077] Considering the nonlinear, asymmetric, and abrupt characteristics of ecological stress response mechanisms, this invention employs radial basis function neural networks (RBFNN) as a nonlinear modeling tool to establish a mapping relationship between perturbation feature vectors and stress response indices.

[0078] Model structure: RBFNN consists of an input layer, hidden layers, and an output layer: Input layer: Receives m-dimensional perturbation vectors; Hidden layer: contains h Gaussian kernel function nodes, each node center represents a perturbation mode; Output layer: Outputs the ecological degradation risk value Rv, ranging from [0,1].

[0079] Kernel function settings: Using Gaussian function , where: x is the input perturbation vector; c is the center vector of the node; γ is the width coefficient, representing the response sensitivity, with a preferred range of [0.01, 0.2]; all node parameters are trained through supervised learning.

[0080] Using historical monitoring data samples (N ≥ 100), with the perturbation vector as input and three types of stress indicators as target values, the RBFNN is trained using the following steps: Kernel centers are initialized using K-means clustering. Weighted learning: Optimize the objective function using the minimum mean square error (MSE); Training convergence criterion: If the loss function decreases by less than 1% in 5 consecutive iterations, the model is considered to have converged.

[0081] After training, the mapping function is formed: ,in As weight, Let φ be the core center and φ be the Gaussian response function.

[0082] The degradation risk value Rv, as a continuous variable, ranges from [0,1] and represents the probability and severity of ecological degradation. Each level corresponds to a different level of governance recommendation, which can be combined with the spatial coding of river sections to form a two-dimensional risk distribution map, which can be used by managers for prioritizing ecological restoration and allocating resources.

[0083] The final degenerate function model supports the following output: The Rv value for each river segment; The corresponding degradation level label; Contribution rate of primary stress factors (through weighted backtracking).

[0084] This model can be deployed in ecological monitoring platforms, watershed management systems, or urban water environment intelligent analysis platforms to achieve real-time degradation early warning and auxiliary decision-making for governance strategies.

[0085] After constructing the ecological degradation risk function and calculating the risk value, this invention further proposes a mutation segmentation threshold method to divide the risk value sequence into five degradation levels: mild, mild-moderate, moderate, severe, and extremely severe. Unlike traditional equidistant segmentation or manual experience-based segmentation, this method automatically identifies and segments based on the mutation characteristics of the risk value itself, ensuring that the segmentation results truly reflect the degradation and evolution process of the river ecosystem.

[0086] The input data is the sequence of ecological degradation risk values ​​obtained in the previous step, denoted as: ;in: This represents the degradation risk value of the u-th sample (time point or spatial location); u represents the total number of samples; each risk value ranges from [0,1], and the closer it is to 1, the higher the risk of ecological degradation.

[0087] To achieve automated segmentation based on risk values, this invention introduces a sliding window first-order difference method combined with a standard deviation multiple threshold method to identify abrupt change points in risk values.

[0088] Set the window length to w (preferably 5 to 7), and perform local first-order differencing on the risk value sequence: ; where ΔRvi represents the magnitude of risk value change between two adjacent samples at position i.

[0089] To avoid the scaling effect caused by different sequence lengths, the difference values ​​are normalized to the rate of change: ; where Rv_max and Rv_min are the maximum and minimum values ​​in the risk value sequence, respectively.

[0090] The mutation threshold λ is set using the standard deviation multiple method: Where μΔ represents the average of all ΔRateᵢ; σΔ represents the standard deviation of all ΔRateᵢ; α is a multiplier, preferably ranging from 2 to 3, and is preferably set to 2.5 in this invention. When ΔRatei ≥ λ, position i is determined to be the mutation point of the risk value sequence.

[0091] After identifying the mutation points, the following steps are used to classify the levels: The risk value sequence is divided into several intervals using the mutation point as the boundary. The segment set is defined as: S = {S1, S2,…, Sj}; where each interval Sj contains a set of risk value samples.

[0092] For each interval Sj, calculate its mean μj and boundary values ​​(minimum and maximum). Where |Sj| represents the number of samples contained in the interval.

[0093] Arrange the boundary values ​​of each interval in ascending order to obtain a set of threshold values: ; where θj represents the dividing point of the j-th level.

[0094] Based on the threshold set Θ, the risk value range is mapped to five degradation levels: Rv ∈ [0, θ1): Slight degradation; Rv ∈ [θ1, θ2): Mild to moderate degradation; Rv ∈ [θ3, θ4): Moderate degradation; Rv ∈ [θ4, θ5): severe degradation; Rv ∈ [θ5, 1]: Extremely severe degradation.

[0095] To improve the applicability of diagnostic results, this invention further proposes to combine the grading results with the spatial coordinates of the river section to generate a two-dimensional grading distribution map.

[0096] Each risk value sample Rvi is bound to the coordinates (xi, yi) of the sampling point or monitoring section to form a triple (xi, yi, Li), where Li represents the corresponding level result.

[0097] The inverse distance weighting (IDW) method is used to generate a graded distribution surface across the entire river section. The calculation formula is as follows: Where: di represents the distance between position (x,y) and sample point (xi, yi); p is the power exponent, preferably 2; L(x,y) represents the degradation level of the predicted position.

[0098] Five levels of regions are distinguished by different color gradients, generating a two-dimensional spatial level distribution map to visually demonstrate the spatial differences in ecological degradation.

[0099] After completing the calculation and classification of ecological degradation risk values, this invention further proposes a method combining reverse feature projection and hot spot clustering identification to extract the spatial coordinate set of ecologically weak areas from river sections with severe and extremely severe degradation levels.

[0100] Input data includes: the calculated sequence of ecological degradation risk values ​​Rv = {Rv1, Rv2, …, Rvn}; and the main perturbation feature subspace. , where each vector Indicates the direction of the main disturbance; the coordinate set of river section sampling points or monitoring sections P = {(x1, y1), (x2, y2), …, (x n, y n )}.

[0101] in This represents the degradation risk value of the i-th sampling point, ranging from [0,1]; if If the value is ≥ 0.60, it is determined to be a severe or extremely severe degradation point and proceeds to the next analysis; there is a one-to-one correspondence between C and Rv, that is, each risk value corresponds to a principal disturbance vector.

[0102] The purpose of inverse feature projection is to map the risk value back to the main perturbation feature space to obtain the performance of that point on the multidimensional perturbation factor.

[0103] Based on the results of higher-order singular value decomposition, the ecological feature tensor T has been decomposed into the core tensor G and orthogonal matrices U1, U2, and U3. Denotes an orthogonal basis along the principal component dimensions; each perturbation eigenvector Indicates a main disturbance direction; and A mapping relationship exists. The back projection calculation method is as follows: ;in: This is the feature distribution vector after back projection; This represents the tensor product in the second dimension; This is the risk value normalization function, used to extend Rvᵢ to the weight space corresponding to the perturbation direction.

[0104] Back projection The vector preserves the correspondence between risk values ​​and disturbance factors, and can reveal the "main disturbance pattern corresponding to high-risk points".

[0105] In order to identify weak regions with spatial concentration in the feature space, this invention uses the density-based spatial clustering algorithm DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to cluster the feature distribution vector set after back projection.

[0106] The basic idea of ​​DBSCAN is that if the number of sample points in the neighborhood of a point exceeds a set threshold, then that point and its neighborhood constitute a cluster. This method is suitable for discovering irregularly shaped cluster regions and can automatically distinguish noise points.

[0107] Neighborhood radius ε: set to 1.5 times the mean square error σ of all sample features; Minimum number of samples MinPts: set to 5, indicating that at least 5 neighboring points are required to form an effective hotspot; Distance metric: Euclidean distance.

[0108] All back-projected feature vectors Standardize the process; Calculate the number of points in the ε-neighborhood of each point; If the number of points in the neighborhood is greater than or equal to MinPts, then mark the point as a core point; The core point and its neighboring points are merged into a cluster; Repeat the above steps until all points are classified or labeled as noise.

[0109] Obtain the cluster set Each cluster represents a potential degraded hot spot region, and v is the total number of clusters.

[0110] To map feature clusters to spatial locations, this invention proposes the following method: Each back-projected feature point All correspond to the coordinates of the sampling points The set of coordinate points belonging to the same cluster is denoted as: .

[0111] For each set of coordinates The minimum bounding rectangle fitting method is used to obtain the spatial coverage boundary of the cluster. The rectangular boundary is determined by the coordinates of its four vertices. The geometric center of the rectangular boundary is calculated as the representative coordinate point of the weak region: .

[0112] The final output is the coordinate set of ecologically degraded and vulnerable areas: .

[0113] To achieve a quantitative explanation of the causes of degradation, this invention constructs a feature source tracing matrix M based on the main perturbation feature vector set C and the weight parameters of the risk function.

[0114] Input is the main perturbation feature vector ; For each perturbation vector, backtrack its weighted contribution value in the risk function; The contribution values ​​are categorized and matrixed according to four parameter dimensions: physicochemical, biological, structural and functional. The elements of the final matrix M This represents the contribution rate of the i-th type of parameter to the j-th degradation factor. Matrix M not only reveals the main driving factors of degradation but also quantifies the strength of the role of each type of factor in the overall degradation.

[0115] Spatial degradation trend maps are used to reflect the distribution of degradation at different levels and their evolution trends.

[0116] First, the degradation level of each monitoring section or sampling point is mapped to spatial coordinates; Two-dimensional hierarchical distribution surfaces are generated using the inverse distance weighted interpolation (IDW) method; To reveal the changing trends over time, results from multiple monitoring periods can be overlaid and compared to form a dynamic evolution layer.

[0117] The generated trend map is marked with a graded coloring method, from light green (excellent) to dark red (extremely severe), which intuitively reflects the trend of degradation aggravation from upstream to downstream in the spatial dimension of the river section, and clarifies the spatiotemporal evolution law of the degradation hot zone.

[0118] Combining the attribution matrix and spatial trend map, this invention further generates intervention priority suggestion paths, specifically including: The coordinate sets of vulnerable areas are sorted, with priority based on the degradation level and the comprehensive weight of driving factors. If the main driving factor is physicochemical parameters, it is recommended to control pollutant input at the source. If the driving factor is structural damage, it is recommended to carry out riparian zone restoration or cross-section reshaping. If the driving factor is functional disorder, it is recommended to improve system stability through ecological water replenishment or biological reconstruction. The above intervention recommendations are output as a path table, which includes the region number, main problems, recommended measures, and priority order.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multidimensional diagnostic method for the degree of river ecological degradation, characterized in that: include: Obtain the basic ecological parameter set of the target river section, including the multiphase physicochemical coupling parameter set P′, the variable biological community response factor set B′, the multiscale structural instability index set H′, and the ecological function coupling parameter set F′; P′, B′, H′ and F′ are respectively subjected to wavelet transform and principal component merging to construct the multidimensional ecological feature tensor T of the target river segment; Using the multidimensional tensor decomposition algorithm, the dominant perturbation feature subspace C in tensor T is extracted to obtain the set of dominant perturbation feature vectors. n is the total number of principal perturbation eigenvectors; Construct a nonlinear degradation function based on the ecological stress response mechanism and output the ecological degradation risk value of the target river section; Based on the abrupt change points of the ecological degradation risk value, the ecological degradation level is divided into 5 levels using the abrupt change segment threshold method: mild, mild-moderate, moderate, severe, and extremely severe. The river sections with severe and extremely severe degradation levels were subjected to reverse feature projection, and hot spot clustering was performed based on the distribution of main disturbance features to obtain the coordinate set of ecologically weak areas. Output a multi-dimensional ecological degradation diagnostic report. This includes a feature source tracing matrix, a spatial degradation trend map, and recommended intervention priority paths.

2. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 1, characterized in that: The set of basic ecological parameters includes: a multiphase physicochemical coupling parameter set P′, including water temperature-dissolved oxygen co-response curves at multiple points within the same period, dynamic ratios of nitrogen and phosphorus, and turbidity-particulate matter correlation factors; a set of variable biological community response factors B′, including watershed-specific taxa response indices, the ratio of pollution-sensitive to pollution-tolerant species, and microbial community Beta diversity indicators; a set of multiscale structural instability indicators H′, including riverbank damage and fracture density, tributary inflow-confluence disturbance index, and cross-sectional hydrodynamic asymmetry coefficient; and a set of ecological function coupling parameters F′, including primary productivity variation rate, bioenergy conversion efficiency, and nitrogen cycle coupling index.

3. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 2, characterized in that: The multidimensional ecological feature tensor T of the target river section includes: The wavelet feature vector group is input into the principal component analysis model, and the top k principal component vectors with a cumulative contribution rate greater than 90% are retained; the loading coefficients of the principal component vectors are used to quantify the contribution of each original parameter in the principal component; the principal component vectors are normalized to obtain the ecological disturbance principal vector set under a unified scale. The merged principal vector set is represented by a matrix to correspond to the four parameter sets, and four principal component sub-matrices MP, MB, MH, and MF are constructed with dimensions m×k, where m is the number of monitored samples and k is the number of principal components. The principal component submatrices MP, MB, MH, and MF are concatenated into tensors with dimension alignment; a third-order tensor T∈ is constructed using tensor stacking operations. The third dimension is used to distinguish the principal component expressions of the four types of parameter sets.

4. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 3, characterized in that: The extraction of the dominant perturbation feature subspace C in tensor T includes: Perform high-order singular value decomposition (HOSVD) on the constructed multidimensional ecological feature tensor T; The tensor T is decomposed into a core tensor G and three orthogonal matrices U1, U2, and U3, which represent the orthogonal feature spaces of the sample dimension, principal component dimension, and feature class dimension, respectively. The core tensor G is used to characterize the joint contribution strength of each perturbation factor, and the column vectors in the orthogonal matrix U2 are the main perturbation eigenvector set C; The total number of vectors n in subspace C is determined based on the principle that the cumulative contribution rate reaches 95%.

5. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 4, characterized in that: The HOSVD decomposition process employs a truncation strategy, including: Calculate the nth singular values ​​of tensor T in each dimension and sort them in descending order; Select the feature vectors corresponding to the first r singular values ​​of each dimension to construct a subspace, where r is the minimum dimension required to satisfy the cumulative contribution rate not being lower than a set threshold. The core tensor G is compressed and preserved, retaining only the tensor block corresponding to the r-dimensional dimension, and removing perturbation factors with low contribution.

6. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 1, characterized in that: The construction of the nonlinear degradation function based on the ecological stress response mechanism includes: Select all or part of the vectors in the main perturbation feature subspace C as input terms of independent variables; Construct an ecological stress response index set R, including the abundance of mutant groups, functional loss factors, and the frequency of redox potential extreme values; A nonlinear regression modeling strategy is adopted to fit the mapping relationship between feature vectors and stress response indicators; The fitting function, R = f(C), is used to predict the ecological degradation risk value Rv of the target river section.

7. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 6, characterized in that: The classification based on abrupt changes in ecological degradation risk values ​​includes: The ecological degradation risk value sequence of the target river section is sorted according to time or spatial dimensions; The rate of change of adjacent risk values ​​is calculated using the sliding window first-order difference method. When the rate of change exceeds the set threshold λ, the location is determined to be a risk value mutation point; Using all mutation points as interval dividing nodes, the risk value interval is divided into several segments; Arrange the segments after dividing the risk value range in numerical order; Calculate the mean and boundary points of each segment interval; Using the boundary values ​​of the segmented intervals as the threshold for classifying the severity, five severity levels are obtained: mild, mild-moderate, moderate, severe, and extremely severe.

8. The multidimensional diagnostic method for the degree of river ecological degradation according to claim 7, characterized in that: Inverse feature projection, including: A mapping relationship is established between the risk values ​​corresponding to the river sections identified as severe and extremely severe and their main disturbance feature vectors; Using the core tensor G and orthogonal matrix The reverse product method maps the risk value back to the main perturbation feature subspace; The characteristic distribution vector of the river segment on the multidimensional perturbation factor is obtained.

9. A multidimensional diagnostic method for the degree of river ecological degradation according to claim 8, characterized in that: The hotspot clustering identification based on the distribution of principal perturbation features includes: Perform standardization on the feature distribution vector set obtained by back projection; The density clustering algorithm DBSCAN is used to identify high-density feature regions. The neighborhood radius ε is set to 1.5 times the mean square error of the sample features, and the minimum number of samples MinPts is 5. The distribution points identified as core samples and their neighborhoods are marked as potential hot spot regions. Spatial localization methods for potential hot spot regions include: extracting the coordinates of the monitoring section corresponding to each feature distribution point; By performing minimum bounding rectangle fitting on the coordinate set within the same cluster, the spatial boundary of the degenerate region can be obtained. The coordinates of the center point of the output boundary are used as the coordinate set of ecologically degraded and vulnerable areas.

Images