A Comprehensive Assessment Method and System for Biodiversity Impact Based on Multiple Ecological Factors

Abstract

This invention provides a method and system for comprehensive assessment of biodiversity impacts based on multi-source ecological factors, relating to the field of biodiversity analysis technology. The method includes acquiring multi-source ecological factor data of areas affected by geological disaster control, performing maximum likelihood estimation and kernel function learning to obtain a spatiotemporal intensity field; performing adaptive time-frequency decomposition processing based on the spatiotemporal intensity field and the time series of community functional traits to obtain a multi-scale mode set; performing low-rank dynamic identification processing on the multi-scale mode set to obtain approximate operator and control gain pairs; performing sparse regression modeling to obtain a set of nonlinear state equations; and performing distribution displacement measurement processing based on the set of nonlinear state equations and a preset quadrat baseline distribution to obtain a comprehensive assessment result of biodiversity impacts. This invention overcomes the problems of single indicators, coarse modeling, and lack of multi-scenario prediction capabilities in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of biodiversity analysis technology, and more specifically, to a method and system for comprehensive assessment of biodiversity impacts based on multiple ecological factors. Background Technology

[0002] Currently, the disturbance of regional ecosystems by geological disaster management is unavoidable, and its impact on biodiversity is receiving increasing attention. Existing technologies for ecological impact assessment typically rely on single or limited ecological factors, such as changes in vegetation cover, species abundance statistics, or local hydrological monitoring indicators, mainly using static monitoring or empirical models for qualitative or semi-quantitative analysis. While these methods can reflect the trends of ecosystems after disturbance to some extent, they often neglect the dynamic coupling relationships and spatiotemporal evolution characteristics among multiple ecological factors. Furthermore, traditional methods often employ modeling techniques based on experience or linear assumptions, lacking the ability to accurately characterize the triggering effects of disturbances and failing to reveal the complex nonlinear response processes of post-disaster community functional traits.

[0003] Therefore, in complex disturbance scenarios, existing technologies often suffer from insufficient accuracy, poor adaptability, and weak interpretability, making it difficult to meet the practical needs of current ecological restoration and biodiversity conservation. There is an urgent need for a comprehensive assessment method and system for biodiversity impacts based on multiple ecological factors to address these technical problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for comprehensive assessment of biodiversity impacts based on multiple ecological factors, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a comprehensive assessment method for biodiversity impacts based on multiple ecological factors, including:

[0006] Acquire multi-source ecological factor data of the area affected by geological disaster control, including spatiotemporal event sequences of construction and natural disturbances, time series of community functional traits, and hydrological observation sequences;

[0007] The multi-source ecological factor data are processed by maximum likelihood estimation and kernel function learning to obtain the spatiotemporal intensity field that characterizes the disturbance triggering effect.

[0008] Adaptive time-frequency decomposition processing is performed based on the time series of the spatiotemporal intensity field and the community functional traits to obtain a multi-scale modal set of post-disaster ecological response.

[0009] The multi-scale mode set is subjected to low-rank dynamic identification processing to obtain an approximate operator and control gain pair for the ecological response;

[0010] Based on the multi-scale modal set, the approximate operator of ecological response, and the control gain pair, sparse regression modeling is performed to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

[0011] Based on the set of nonlinear state equations and the preset quadrat baseline distribution, distribution displacement measurement is performed to obtain a comprehensive assessment result of the impact on biodiversity.

[0012] Secondly, this application also provides a comprehensive assessment system for biodiversity impacts based on multiple ecological factors, including:

[0013] The acquisition unit is used to acquire multi-source ecological factor data of the area affected by geological disaster control. The multi-source ecological factor data includes the spatiotemporal event sequence of construction and natural disturbance, the time series of community functional traits, and the hydrological observation sequence.

[0014] The processing unit is used to perform maximum likelihood estimation and kernel function learning on the multi-source ecological factor data to obtain a spatiotemporal intensity field that characterizes the disturbance triggering effect.

[0015] The decomposition unit is used to perform adaptive time-frequency decomposition processing based on the time series of the spatiotemporal intensity field and the community functional traits to obtain a multi-scale modal set of post-disaster ecological response.

[0016] The identification unit is used to perform low-rank dynamic identification processing on the multi-scale mode set to obtain an approximate operator and control gain pair for the ecological response;

[0017] The modeling unit is used to perform sparse regression modeling based on the multi-scale modal set, the approximate operator of the ecological response, and the control gain pair to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

[0018] The evaluation unit is used to perform distribution displacement measurement processing based on the set of nonlinear state equations and the preset quadrat baseline distribution to obtain a comprehensive evaluation result of the impact on biodiversity.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention systematically constructs a nonlinear coupling relationship between disturbance-triggered effects and community dynamic responses by introducing multiple techniques, including maximum likelihood estimation, kernel function learning, adaptive time-frequency decomposition, low-rank dynamics identification, and sparse regression modeling. This method not only extracts multi-scale response modes of post-disaster communities from multi-source spatiotemporal events caused by construction and natural disturbances, but also simulates community evolution trajectories under different disturbance scenarios through a set of nonlinear state equations. Finally, it combines an optimal transport model to achieve a quantitative assessment of the impact on biodiversity, thus overcoming the shortcomings of existing technologies such as single indicators, coarse modeling, and lack of multi-scenario prediction capabilities. This provides more scientific, accurate, and actionable decision support for ecosystem governance and biodiversity conservation.

[0021] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the process for the comprehensive assessment of biodiversity impacts based on multiple ecological factors as described in this embodiment of the invention.

[0024] Figure 2 This is a schematic diagram of the biodiversity impact comprehensive assessment system based on multi-source ecological factors as described in this embodiment of the invention.

[0025] In the diagram: 701, Acquisition Unit; 702, Processing Unit; 703, Decomposition Unit; 704, Recognition Unit; 705, Modeling Unit; 706, Evaluation Unit; 7021, First Processing Subunit; 7022, Second Processing Subunit; 7023, Third Processing Subunit; 7031, First Decomposition Subunit; 7032, Second Decomposition Subunit; 7033, Third Decomposition Subunit; 7041, First Recognition Subunit; 7042, Second Recognition Subunit; 7043, Third Recognition Subunit; 7051, First Modeling Subunit; 7052, Second Modeling Subunit; 7053, Third Modeling Subunit. Detailed Implementation

[0026] 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] Example 1

[0029] This embodiment provides a comprehensive assessment method for biodiversity impact based on multiple ecological factors.

[0030] See the instruction manual appendix Figure 1 The figure shows that the method includes steps S1, S2, S3, S4, S5 and S6.

[0031] Step S1: Obtain multi-source ecological factor data of the area affected by geological disaster control. The multi-source ecological factor data includes the spatiotemporal event sequence of construction and natural disturbance, the time series of community functional traits, and the hydrological observation sequence.

[0032] It is understandable that this step establishes a comprehensive data foundation through system integration and multi-dimensional dynamic monitoring technology. Spatially, it utilizes high-resolution UAV imagery (0.1m accuracy). 2 The system combines a grid-based approach with satellite remote sensing to precisely locate the scope of construction disturbances (such as spoil heap excavation and road construction) and natural geological hazard points (such as landslide boundaries). RTK mapping equipment is used to georeference elevation layers (contour lines every 10 meters) with the micro-topography of the travertine dam, ensuring spatial accuracy of spatiotemporal events. In the temporal dimension, a network of hydrological sensors (pressure level gauges and turbidity meters) is deployed to collect minute-level continuous sequences of runoff and sediment content in real time. Simultaneously, seasonal dynamic data on community functional traits (such as leaf nitrogen content, specific leaf area, and canopy height) are obtained through fixed sampling methods (once per quarter) and remote sensing vegetation indices.

[0033] Step S2: Perform maximum likelihood estimation and kernel function learning on the multi-source ecological factor data to obtain the spatiotemporal intensity field characterizing the disturbance triggering effect;

[0034] It is understandable that this step transforms multi-source heterogeneous ecological data into a spatiotemporal intensity field with quantifiable perturbation effects. Essentially, it constructs a three-dimensional probability density field describing the "probability of perturbation event occurrence - spatial location - time" through statistical learning. In this step, step S2 includes steps S21, S22, and S23.

[0035] Step S21: Perform spatiotemporal event construction processing on the multi-source ecological factor data. By identifying the mutation time point for each data stream, and taking the preset elevation stratification and travertine dam location mutation as a single trigger event, a spatiotemporal event sequence is obtained. The time and location accuracy of each event are corrected by the preset UAV image grid.

[0036] Understandably, this step constructs a high-precision spatiotemporal event sequence by identifying abrupt changes in time points from multi-source data streams and constraining them with geospatial rules. First, abrupt changes in time points are identified for three types of data streams: construction disturbances (e.g., mechanical vibration waveforms), natural changes (e.g., hydrological parameter drift), and travertine dam morphology (based on UAV imagery). For construction disturbances, a dynamic threshold method is used to capture sudden amplitude increases (e.g., greater than 30 dB). For hydrological parameters, a sliding window variance analysis is used to detect statistical characteristic jumps (e.g., standard deviation > 3 times the baseline within 24 hours). For travertine dams, morphological contour analysis is used to identify dam boundary displacements (preset displacements). The event sequence is designed to detect disturbances in travertine terraces (e.g., a depth > 2 cm). Simultaneously, based on pre-defined elevation stratification rules (e.g., 1500-1600m is designated as a shrub-grass transition zone), all concurrent disturbances within this layer (e.g., a sudden increase in vibration + a sharp drop in vegetation area) are forcibly merged into a single event. Furthermore, the ecological response is correlated through abrupt changes in the location of the travertine terrace dam (e.g., a porosity decrease > 10% triggering dam instability). Finally, using UAV image grid correction technology, the event location is projected onto a pre-defined geographic grid (e.g., 1m × 1m) and the timestamp is calibrated to eliminate terrain distortion and clock asynchrony errors, outputting an event sequence with spatiotemporal accuracy of ±0.8m and ±2 seconds. This approach successfully decouples construction blasting from aftershock disturbances during post-earthquake repair and reveals the specific propagation pattern of disturbances along the 1600m contour line through elevation stratification rules, providing error-free event input for the spatiotemporal intensity field.

[0037] Step S22: Perform preliminary fitting processing based on maximum likelihood estimation according to the spatiotemporal event sequence to obtain preliminary parameterized kernel coefficients and spatial background field for nonparametric kernel learning;

[0038] It is understandable that in this step, by constructing a spatiotemporal point process model and implementing maximum likelihood estimation (MLE) optimization, the initial parameters required for nonparametric kernel learning (preliminary parameterized kernel coefficients and spatial background field) are output. Its mathematical model is defined as:

[0039] ,

[0040] in, λ (x, y, t) is the spatiotemporal intensity field function, describing the perturbation triggering intensity at location (x, y) and time t; μ(x, y) is the spatial background field function, describing the fundamental perturbation level at location (x, y) and time t; k is the parameterized kernel function; t j Let x be the time when the j-th perturbation event occurs. j and y j All are spatial coordinates of the j-th perturbation event, x and y are the geographical locations of the previous perturbation event, j is the j-th perturbation event, and t is the time of occurrence of the previous perturbation event.

[0041] Log-likelihood function: ,

[0042] Where log L is the objective function (log-likelihood function). Let x be the intensity function, describing the event at position x. j， y j and time t i Let i represent the instantaneous probability density at point i, where i represents the i-th perturbation event, and n represents the total number of perturbation events. This represents the triple integral over the spacetime infinitesimal element dxdydt. λ The intensity of the spatiotemporal impact of construction blasting.

[0043] This step employs the gradient ascent method to iteratively optimize the log-likelihood function, solving for the rasterized distribution of the spatial background field μ(x,y). Then, the coefficients of the parameterized kernel function κ are simultaneously fitted to obtain the initial parameters. The kernel function used in this step is an exponentially decaying kernel, and its formula is shown below: ;

[0044] in, Let α be the kernel function, Δt be the time difference, Δr be the spatial distance, α be the basic intensity coefficient, β be the time decay coefficient, and γ be the spatial decay coefficient.

[0045] Step S23: Based on the preliminary parameterized kernel coefficients and the spatial background field, perform nonparametric learning and uncertainty quantification of the kernel function. By constructing a spacetime basis function and introducing time monotonicity and nonnegativity constraints, a spacetime intensity field characterizing the disturbance triggering effect is obtained.

[0046] Understandably, this step is based on the preliminary parameterized kernel coefficients and spatial background field. By constructing a spacetime basis function and introducing time monotonicity and non-negativity constraints, high-fidelity modeling of the perturbation triggering effect is achieved: First, the kernel function is decomposed into the tensor product form of the time basis function (discrete trigonometric function describing the decay characteristics) and the spatial basis function (radial basis function describing the diffusion characteristics). Then, the negative log-likelihood function is minimized, and time monotonicity and non-negativity constraints are applied.

[0047] The time monotonicity constraint is as follows: ,

[0048] Where k is the time basis function index, m is the space basis function index, and C km This represents the interaction strength between the time basis functions and the space basis functions. Ψm represents the rate of change of the time decay rate, and represents the diffusion pattern of the disturbance effect with spatial distance (Gaussian diffusion). This indicates that the constraint condition holds for all time differences Δt and spatial distances Δr.

[0049] The nonnegativity constraints are as follows: ,

[0050] Among them, C km Let k represent the time basis function and m be the index of the space basis function. The constraint condition holds for all time basis functions and space basis functions.

[0051] Meanwhile, this step uses Bayesian posterior sampling to generate a parameter set and calculates the confidence interval of the spatiotemporal intensity field. This step reduces the error of the traditional Gaussian kernel and avoids redundant ecological compensation area.

[0052] Step S3: Based on the time series of the spatiotemporal intensity field and the community functional traits, perform adaptive time-frequency decomposition processing to obtain a multi-scale modal set of post-disaster ecological response;

[0053] Understandably, this step reveals the multi-scale response mechanism of post-disaster ecological disturbance through joint time-frequency domain analysis: First, using the spatiotemporal intensity field as a guiding signal, synchronous compressed wavelet transform is performed on the time series of community functional traits (such as canopy height and chlorophyll content)—multi-resolution analysis is performed using complex Morlet wavelet kernels, and instantaneous frequency compression energy diffusion is calculated through phase gradient; based on the time-varying characteristics of the intensity field, the filtering bandwidth is dynamically adjusted (e.g., the high-frequency band is broadened to 0.1-1Hz after the blasting event), and the dominant modes are screened by combining the energy entropy criterion (threshold He>0.85), eliminating noise components (such as random fluctuations caused by instrument errors); finally, high-frequency oscillations that are strictly aligned with the disturbance event (such as instantaneous stomatal closure triggered by construction vibration), mid-frequency fluctuations (such as the biomass cycle changes affected by monsoon precipitation), and low-frequency trends (such as the community succession direction caused by travertine degradation) are separated, constructing a physically interpretable multi-scale mode set. In this step, step S3 includes steps S31, S32, and S33.

[0054] Step S31: Perform multi-resolution wavelet pre-decomposition processing based on the spatiotemporal intensity field and community functional trait time series, extract the disturbance response components of different frequency bands through continuous wavelet transform, and redistribute the time-frequency energy using the synchronous compression method to obtain a preliminary time-frequency localization representation.

[0055] Understandably, this step utilizes multi-resolution wavelet pre-decomposition and synchronous compression techniques to construct a refined representation of the ecological response in the time-frequency domain. First, the spatiotemporal intensity field is used as the perturbation-driven signal input, and a continuous wavelet transform is performed synchronously with the time series of community functional traits (such as photosynthetic rate and root biomass). The original signal is decomposed at multiple scales using a complex Morlet wavelet basis (matching plant physiological cycles) to generate an initial time-frequency coefficient matrix, whose scale parameters cover the 0.01–10Hz ecological response frequency band in a logarithmic distribution. To address energy diffusion defects (such as the blurring of the blasting perturbation signal in the 2–3Hz band), this step introduces synchronous compression transform (SST) to recalibrate the time-frequency plane: by calculating the instantaneous frequency of the analytical wavelet coefficients, the diffused energy is recompressed to the true frequency ridge, generating a time-frequency distribution with focused energy. The instantaneous frequencies of the wavelet coefficients are shown below: ,

[0056] Where, ω f (a,b) represents the local frequency response of the signal at scale a and time shift b, where q represents the imaginary unit, and W... f (a,b) are wavelet coefficients. The symbol represents the partial derivative.

[0057] Step S32: Perform empirical wavelet decomposition based on the time-frequency localization representation, separate the dominant mode of the community functional trait sequence from the background noise by adaptively constructing a filter function, and filter the effective modes by combining the energy entropy criterion to obtain a noise-reduced multi-scale mode subset;

[0058] Understandably, this step separates the dominant mode of the ecological response based on time-frequency localization representation and adaptive empirical wavelet decomposition techniques: First, an adaptive filter function group is constructed—the frequency band boundaries are automatically divided according to the energy distribution spectrum (e.g., in the Jiuzhaigou case, a bandpass boundary is set at the dominant frequency of the blasting disturbance at 2.5Hz, and the energy ridge inflection point is detected through the frequency band segmentation threshold (avoiding the loss of weak signals due to manually preset frequency bands), generating a compactly supported orthogonal wavelet filter group, where the filters are shown below:

[0059] ,

[0060] in, Let z be the filter function for the nth frequency band, z be the roll-off shape function, and γ be the filter function for the nth frequency band. n Where ω is the half-width of the transition band, and ω is the angular frequency variable. The left boundary frequency of the nth frequency band. It is the right boundary frequency of the nth frequency band.

[0061] This step extracts modal components by convolving the time-frequency representation with the filter bank, and then filters effective modes based on the energy entropy criterion, eliminating background noise components. The adaptive detection mechanism of the mid-frequency band boundary (whether it is greater than a preset threshold) in this step solves the defect of traditional wavelet decomposition relying on a preset scale. Combined with the energy entropy criterion, it realizes intelligent discrimination of noise and weak ecological response, which improves the effective mode capture rate compared with the fixed threshold method.

[0062] Step S33: Perform Hilbert spectral analysis based on the multi-scale modal subset, and characterize the dynamic response features of the community at different time scales after disturbance by calculating the instantaneous frequency and instantaneous energy curves, thereby obtaining the multi-scale modal set of post-disaster ecological response.

[0063] Understandably, this step is based on a denoised multi-scale modal subset and uses Hilbert spectral analysis to quantify the dynamic response characteristics of the post-disaster community. This step first performs a feature extraction transformation with spatiotemporal intensity field constraints on each mode—using the Hilbert transform to extract the instantaneous frequency and energy envelope of the oscillation mode. Specifically, the instantaneous frequency is time-varyingly weighted by integrating the perturbation gradient (e.g., giving high weight to the area within a 5-minute window after the blasting event), and a spatial gradient decay factor is applied to the instantaneous energy to suppress random fluctuations unrelated to the perturbation (e.g., instrument noise). This step injects the spatiotemporal pattern of the perturbation into the instantaneous attribute analysis (e.g., giving higher weight to the area surrounding the blasting point), addressing the shortcomings of traditional methods in characterizing coupling effects.

[0064] Step S4: Perform low-rank dynamic identification processing on the multi-scale mode set to obtain the approximate operator and control gain pair of the ecological response;

[0065] It is understandable that this step extends the modal time series to a high-dimensional phase space based on the Hankel matrix, extracts the dominant eigenvectors using singular value decomposition (SVD), removes noise interference (retaining 10-dimensional principal components with a variance contribution rate >95%), forms low-rank state variables, and then introduces the spatiotemporal intensity field as an external control input. Finally, the state transition matrix and control gain matrix are solved by dynamic mode decomposition (DMD), and then the ridge regression algorithm is used to suppress overfitting and ensure the sparsity of the transition matrix and gain matrix. In this step, step S4 includes steps S41, S42 and S43.

[0066] Step S41: Perform time extension embedding processing on the multi-scale mode set, and extract the dominant modes of the multi-scale mode set by constructing the Hankel matrix and using singular value decomposition to obtain a low-rank state space representation;

[0067] Understandably, this step reconstructs the ecological dynamics phase space through time-extended embedding: A Hankel state matrix is ​​constructed by taking a multi-scale modal set (such as the high-frequency mode of moss photosynthetic inhibition and the low-frequency mode of cyanobacterial succession) with a preset time lag (determined based on the first zero-crossing point of the autocorrelation function). Its row vectors are formed by linear extension of the sampling points of each mode within a continuous time window. Then, noise interference is removed based on singular value decomposition (SVD)—retaining principal components with a cumulative contribution rate > 95%, generating low-rank state variables that characterize the essential dynamics of the system. This step can compress the original 11-dimensional noise modes into 2-dimensional state variables. This step fully preserves the disturbance-response causal mechanism, laying the foundation for control gain modeling.

[0068] Step S42: Perform dynamic mode decomposition processing based on the low-rank state space representation. Decompose the state transition matrix using the least squares method and introduce external disturbances as control variables to obtain the Koopman operator and initial value of control gain in the low-rank state space representation.

[0069] Understandably, this step, based on low-rank state-space representation, constructs an initial dynamic model of the ecosystem using control-enhanced dynamic mode decomposition (DEDM). First, augmented matrices are constructed from the state variables and their time derivatives (introducing a spatiotemporal intensity field as an external control input to construct a least-squares problem). Then, QR decomposition and Moore-Penrose pseudo-inverse are used to solve for the state transition matrix and control gain matrix. In this step, the control-enhanced least-squares architecture treats the intensity field as an independent control term, addressing the distortion problem in ecosystem modeling.

[0070] Step S43: Perform regularization optimization based on the approximate operator and the initial value of the control gain. Use the ridge regression method to suppress noise interference and maintain the sparsity of the system to obtain the approximate operator and control gain pair for the ecological response.

[0071] Understandably, this step addresses the initial values ​​of the dynamic mode decomposition (state transition matrix and control gain) by using a ridge regression regularization algorithm to improve the robustness of the ecological dynamics model. During the solution process, this is achieved by constraining the coefficients of non-critical terms to approach zero (e.g., the weak coupling term between travertine porosity and insect communities is compressed to the order of 10⁻⁶), while preserving core ecological mechanisms (e.g., maintaining the strong control gain of blasting vibration on photosynthesis). This resolves the issues of noise sensitivity and ecological interpretability of the initial value model.

[0072] Step S5: Based on the multi-scale modal set, the approximate operator of ecological response and the control gain pair, perform sparse regression modeling to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

[0073] Understandably, this step constructs a nonlinear ecological dynamics model through sparse regression modeling: First, a feature dictionary matrix is ​​constructed based on a multi-scale modal set (high-frequency mode of photosynthetic inhibition, low-frequency mode of succession trend) and control gain pairs; a minimum angle regression algorithm combined with L1 regularization (with a preset weight of 0.01) is used to screen significant ecological process terms (e.g., retaining only the interaction term between photosynthetic inhibition and blasting intensity), and a Bayesian information criterion is introduced to penalize complexity; finally, a set of nonlinear state equations is generated through ridge regression stability parameter estimation to predict the community evolution trajectory under different disturbance scenarios (e.g., the change in moss recovery cycle when the blasting intensity is ±20%). In this step, step S5 includes steps S51, S52, and S53.

[0074] Step S51: Based on the multi-scale mode set and control gain pair, perform feature dictionary construction processing to obtain the feature dictionary matrix of community state variables;

[0075] It is understandable that this step first takes each component of the low-rank state space representation (such as photosynthetic inhibition intensity and cyanobacteria proportion) as the core independent variable; secondly, based on prior knowledge of ecological mechanisms, a set of nonlinear functions of state variables is constructed; then, the control gain matrix is ​​combined with the spatiotemporal intensity field to generate perturbation-state interaction terms; next, the instantaneous energy of each mode is introduced as a weighting factor to construct weighted features; finally, the feature dictionary matrix is ​​integrated, and its column vectors correspond to potential ecological process terms.

[0076] Step S52: Based on the feature dictionary matrix, the approximate operator of the ecological response and the control gain pair, perform sparse regression modeling. Select significant nonlinear terms by combining the minimum angle regression algorithm with L1 regularization constraints, and introduce Bayesian information criteria to suppress overfitting, thereby obtaining a sparse nonlinear dynamic expression.

[0077] Understandably, this step first uses the ecological response approximation operator as the baseline objective, iteratively selects feature basis functions along the least-angle path using the LARS algorithm, and applies L1 regularization constraints to force the coefficients of non-critical terms (such as the quadratic term of rainfall) to zero; then, for the candidate model set on the LARS path, the optimal complexity model is selected; finally, ridge regression optimization (with a penalty coefficient of 10000) is performed on the selected feature subset to suppress coefficient sensitivity and generate the final nonlinear equation:

[0078] ,

[0079] in, The time derivative of the state vector. This represents the set of filtered feature items. The optimized coupling coefficient, It is a nonlinear characteristic function. λ Λ is the disturbance coefficient. scenario It is the set of perturbation coefficients.

[0080] Step S53: Perform parameter optimization based on the nonlinear dynamic expression, determine the sparsity intensity through cross-validation, and estimate the stability coefficient using ridge regression to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

[0081] Understandably, this step divides the parameters into training and testing sets, optimizes the L1 regularization coefficient with the goal of minimizing the root mean square error of the testing set, selects two core ecological mechanisms (photosynthetic inhibition and succession acceleration), removes noise terms, and then injects perturbation parameters (such as blasting intensity ±20%) to generate a set of nonlinear state equations. The ridge regression algorithm in this step ensures the model remains stable even when environmental factors are highly coupled.

[0082] Step S6: Based on the set of nonlinear state equations and the preset quadrat baseline distribution, perform distribution displacement measurement processing to obtain a comprehensive assessment result of the biodiversity impact.

[0083] Understandably, this step utilizes the set of nonlinear state equations established in the previous steps and combines them with a pre-defined quadrat baseline distribution to measure distribution displacement. It then uses optimal transport theory to characterize the difference between the ecological response distribution under disturbance scenarios and the baseline distribution. This process not only captures the overall shift trend of species composition and functional traits under disturbance, but also comprehensively measures changes in ecological stability and functional diversity at multiple scales, resulting in more accurate and systematic biodiversity impact assessments. This step overcomes the limitations of traditional static index methods, achieving quantitative measurement of multi-scenario, dynamic, and nonlinear community evolution processes, and providing a scientific basis for ecological risk management and restoration strategy formulation.

[0084] The Monge-Kantorovich transport problem is constructed with the pre-defined quadrat baseline distribution (undisturbed state) as the objective: ,

[0085] in, It is a distance metric between probability distributions. To find the minimum transmission cost among all possible coupling methods, Let be a random variable under disturbance. For random variables under the baseline state, For the integral of the coupling measure.

[0086] Then, the Sinkhorn algorithm was used to solve the Wasserstein distance, and weights were assigned to different community functional traits (photosynthetic efficiency, species richness) (determined by entropy weight method) to obtain a comprehensive assessment index of biodiversity.

[0087] Example 2

[0088] As per the instruction manual Figure 2 As shown, this embodiment provides a comprehensive assessment system for biodiversity impact based on multiple ecological factors. See [link to documentation]. Figure 2 The system includes an acquisition unit 701, a processing unit 702, a decomposition unit 703, an identification unit 704, a modeling unit 705, and an evaluation unit 706.

[0089] The acquisition unit 701 is used to acquire multi-source ecological factor data of the area affected by geological disaster control. The multi-source ecological factor data includes the spatiotemporal event sequence of construction and natural disturbance, the time series of community functional traits, and the hydrological observation sequence.

[0090] Processing unit 702 is used to perform maximum likelihood estimation and kernel function learning on the multi-source ecological factor data to obtain a spatiotemporal intensity field that characterizes the disturbance triggering effect;

[0091] The processing unit 702 includes a first processing subunit 7021, a second processing subunit 7022, and a third processing subunit 7023.

[0092] The first processing subunit 7021 is used to perform spatiotemporal event construction processing on the multi-source ecological factor data. By identifying the mutation time point for each data stream, and taking the preset elevation stratification and travertine dam location mutation as a single trigger event, a spatiotemporal event sequence is obtained. The time and location accuracy of each event are corrected by a preset UAV image grid.

[0093] The second processing subunit 7022 is used to perform preliminary fitting processing based on maximum likelihood estimation according to the spatiotemporal event sequence to obtain preliminary parameterized kernel coefficients and spatial background field for nonparametric kernel learning;

[0094] The third processing subunit 7023 is used to perform nonparametric learning and uncertainty quantification of the kernel function based on the preliminary parameterized kernel coefficients and the spatial background field. By constructing a spacetime basis function and introducing time monotonicity and nonnegativity constraints, a spacetime intensity field that characterizes the disturbance triggering effect is obtained.

[0095] Decomposition unit 703 is used to perform adaptive time-frequency decomposition processing based on the time series of the spatiotemporal intensity field and the community functional traits to obtain a multi-scale modal set of post-disaster ecological response.

[0096] The decomposition unit 703 includes a first decomposition subunit 7031, a second decomposition subunit 7032, and a third decomposition subunit 7033.

[0097] The first decomposition subunit 7031 is used to perform multi-resolution wavelet pre-decomposition processing based on the time series of the spatiotemporal intensity field and community functional traits, extract the disturbance response components of different frequency bands through continuous wavelet transform, and redistribute the time-frequency energy using the synchronous compression method to obtain a preliminary time-frequency localization representation.

[0098] The second decomposition subunit 7032 is used to perform empirical wavelet decomposition processing based on the time-frequency localization representation, separate the dominant mode of the community functional trait sequence from the background noise by adaptively constructing a filter function, and filter the effective modes by combining the energy entropy criterion to obtain a noise-reduced multi-scale mode subset.

[0099] The third decomposition subunit 7033 is used to perform Hilbert spectral analysis based on the multi-scale modal subset, and to characterize the dynamic response features of the community at different time scales after the disturbance by calculating the instantaneous frequency and instantaneous energy curves, thereby obtaining the multi-scale modal set of the post-disaster ecological response.

[0100] The identification unit 704 is used to perform low-rank dynamic identification processing on the multi-scale mode set to obtain an approximate operator and control gain pair for the ecological response;

[0101] The identification unit 704 includes a first identification subunit 7041, a second identification subunit 7042, and a third identification subunit 7043;

[0102] The first identification subunit 7041 is used to perform time extension embedding processing based on the multi-scale mode set, and to extract the dominant mode of the multi-scale mode set by constructing a Hankel matrix and using singular value decomposition to obtain a low-rank state space representation.

[0103] The second identification subunit 7042 is used to perform dynamic mode decomposition processing based on the low-rank state space representation, decompose the state transition matrix by least squares method and introduce external disturbances as control variables to obtain the Koopman operator and initial value of control gain in the low-rank state space representation.

[0104] The third identification subunit 7043 is used to perform regularization optimization processing based on the approximate operator and the initial value of the control gain, and to suppress noise interference and maintain the sparsity of the system through the ridge regression method, so as to obtain the approximate operator and control gain pair of ecological response.

[0105] Modeling unit 705 is used to perform sparse regression modeling based on the multi-scale modal set, the approximate operator of ecological response and the control gain pair to obtain a set of nonlinear state equations for generating community evolution trajectories under different disturbance scenarios.

[0106] The modeling unit 705 includes a first modeling subunit 7051, a second modeling subunit 7052, and a third modeling subunit 7053.

[0107] The first modeling subunit 7051 is used to perform feature dictionary construction processing based on the multi-scale mode set and control gain pair to obtain the feature dictionary matrix of community state variables.

[0108] The second modeling subunit 7052 is used to perform sparse regression modeling based on the feature dictionary matrix, the approximate operator of the ecological response and the control gain pair. It selects significant nonlinear terms by combining the minimum angle regression algorithm with L1 regularization constraints and introduces the Bayesian information criterion to suppress overfitting, thereby obtaining a sparse nonlinear dynamic expression.

[0109] The third modeling subunit 7053 is used to perform parameter optimization based on the nonlinear dynamic expression, determine the sparsity intensity through cross-validation, and estimate the stability coefficient using ridge regression to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

[0110] The evaluation unit 706 is used to perform distribution displacement measurement processing based on the set of nonlinear state equations and the preset quadrat baseline distribution to obtain a comprehensive evaluation result of the impact on biodiversity.

[0111] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0112] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A comprehensive assessment method for biodiversity impact based on multiple ecological factors, characterized in that, include: Acquire multi-source ecological factor data of the area affected by geological disaster control, including spatiotemporal event sequences of construction and natural disturbances, time series of community functional traits, and hydrological observation sequences; The multi-source ecological factor data are processed by maximum likelihood estimation and kernel function learning to obtain the spatiotemporal intensity field that characterizes the disturbance triggering effect. Adaptive time-frequency decomposition processing is performed based on the time series of the spatiotemporal intensity field and the community functional traits to obtain a multi-scale modal set of post-disaster ecological response. The multi-scale mode set is subjected to low-rank dynamic identification processing to obtain an approximate operator and control gain pair for the ecological response; Based on the multi-scale modal set, the approximate operator of ecological response, and the control gain pair, sparse regression modeling is performed to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios. Based on the set of nonlinear state equations and the preset quadrat baseline distribution, distribution displacement measurement is performed to obtain a comprehensive assessment result of the impact on biodiversity. The adaptive time-frequency decomposition process based on the time series of the spatiotemporal intensity field and the community functional traits includes: Based on the time series of the spatiotemporal intensity field and community functional traits, multi-resolution wavelet pre-decomposition processing is performed. Perturbation response components of different frequency bands are extracted by continuous wavelet transform, and time-frequency energy is redistributed using a synchronous compression method to obtain a preliminary time-frequency localization representation. Based on the time-frequency localization representation, empirical wavelet decomposition is performed. The dominant mode of the community functional trait sequence is separated from the background noise by adaptively constructing a filter function. The effective mode is then selected by combining the energy entropy criterion to obtain a noise-reduced multi-scale mode subset. Hilbert spectral analysis was performed on the multi-scale modal subset, and the dynamic response characteristics of the community at different time scales after disturbance were characterized by calculating the instantaneous frequency and instantaneous energy curves, thus obtaining the multi-scale modal set of post-disaster ecological response. Specifically, the multi-scale mode set undergoes low-rank dynamic identification processing to obtain approximate operator and control gain pairs for the ecological response, including: Based on the multi-scale mode set, time extension embedding processing is performed. By constructing the Hankel matrix and using singular value decomposition to extract the dominant modes of the multi-scale mode set, a low-rank state space representation is obtained. Dynamic mode decomposition is performed based on the low-rank state space representation. The state transition matrix is ​​decomposed by least squares and external disturbances are introduced as control variables to obtain the Koopman operator and initial value of control gain in the low-rank state space representation. Based on the Koopman operator and the initial value of the control gain, regularization optimization is performed, and noise interference is suppressed and the sparsity of the system is maintained by the ridge regression method to obtain the approximate operator and control gain pair for the ecological response. The sparse regression modeling process based on the multi-scale modal set, the approximate operator of the ecological response, and the control gain includes: Based on the multi-scale mode set and control gain pair, feature dictionary construction is performed to obtain the feature dictionary matrix of community state variables; Based on the feature dictionary matrix, the approximate operator of the ecological response and the control gain pair, sparse regression modeling is performed. The least angle regression algorithm combined with L1 regularization constraint is used to select significant nonlinear terms, and Bayesian information criterion is introduced to suppress overfitting, resulting in a sparse nonlinear dynamic expression. Based on the nonlinear dynamic expression, parameter optimization is performed, sparsity intensity is determined through cross-validation, and stability coefficients are estimated using ridge regression to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

2. The comprehensive assessment method for biodiversity impact based on multiple ecological factors according to claim 1, characterized in that... The multi-source ecological factor data are processed by maximum likelihood estimation and kernel function learning to obtain a spatiotemporal intensity field characterizing the disturbance triggering effect, including: The multi-source ecological factor data is processed to construct spatiotemporal events. By identifying the abrupt change time point for each data stream, and taking the preset elevation stratification and the abrupt change in the location of the travertine dam as a single triggering event, a spatiotemporal event sequence is obtained. The time and location accuracy of each event are corrected by a preset UAV image grid. Based on the spatiotemporal event sequence, a preliminary fitting process based on maximum likelihood estimation is performed to obtain preliminary parameterized kernel coefficients and spatial background field for nonparametric kernel learning; Based on the preliminary parameterized kernel coefficients and the spatial background field, nonparametric learning and uncertainty quantification of the kernel function are performed. By constructing a spacetime basis function and introducing time monotonicity and nonnegativity constraints, a spacetime intensity field characterizing the disturbance triggering effect is obtained.

3. A comprehensive assessment system for biodiversity impact based on multiple ecological factors, characterized in that, include: The acquisition unit is used to acquire multi-source ecological factor data of the area affected by geological disaster control. The multi-source ecological factor data includes the spatiotemporal event sequence of construction and natural disturbance, the time series of community functional traits, and the hydrological observation sequence. The processing unit is used to perform maximum likelihood estimation and kernel function learning on the multi-source ecological factor data to obtain a spatiotemporal intensity field that characterizes the disturbance triggering effect. The decomposition unit is used to perform adaptive time-frequency decomposition processing based on the time series of the spatiotemporal intensity field and the community functional traits to obtain a multi-scale modal set of post-disaster ecological response. The identification unit is used to perform low-rank dynamic identification processing on the multi-scale mode set to obtain an approximate operator and control gain pair for the ecological response; The modeling unit is used to perform sparse regression modeling based on the multi-scale modal set, the approximate operator of the ecological response, and the control gain pair to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios. The evaluation unit is used to perform distribution displacement measurement processing based on the set of nonlinear state equations and the preset quadrat baseline distribution to obtain a comprehensive evaluation result of the impact on biodiversity. The decomposition unit includes: The first decomposition subunit is used to perform multi-resolution wavelet pre-decomposition processing based on the spatiotemporal intensity field and community functional trait time series, extract perturbation response components of different frequency bands through continuous wavelet transform, and redistribute time-frequency energy using synchronous compression method to obtain a preliminary time-frequency localization representation. The second decomposition subunit is used to perform empirical wavelet decomposition processing based on the time-frequency localization representation, and to separate the dominant mode of the community functional trait sequence from the background noise by adaptively constructing a filter function, and to filter the effective modes by combining the energy entropy criterion to obtain a noise-reduced multi-scale mode subset. The third decomposition subunit is used to perform Hilbert spectral analysis based on the multi-scale modal subset. By calculating the instantaneous frequency and instantaneous energy curves, the dynamic response characteristics of the community at different time scales after the disturbance are characterized, and a multi-scale modal set of post-disaster ecological response is obtained. The identification unit includes: The first identification subunit is used to perform time-extended embedding processing based on the multi-scale mode set, and to extract the dominant mode of the multi-scale mode set by constructing a Hankel matrix and using singular value decomposition to obtain a low-rank state space representation. The second identification subunit is used to perform dynamic mode decomposition processing based on the low-rank state space representation. It decomposes the state transition matrix by least squares method and introduces external disturbances as control variables to obtain the Koopman operator and initial value of control gain in the low-rank state space representation. The third identification subunit is used to perform regularization optimization processing based on the Koopman operator and the initial value of the control gain, and to suppress noise interference and maintain the sparsity of the system through the ridge regression method to obtain the approximate operator and control gain pair of the ecological response. The modeling unit includes: The first modeling subunit is used to perform feature dictionary construction processing based on the multi-scale mode set and control gain pair to obtain the feature dictionary matrix of community state variables; The second modeling subunit is used to perform sparse regression modeling based on the feature dictionary matrix, the approximate operator of the ecological response and the control gain pair. It selects significant nonlinear terms by combining the minimum angle regression algorithm with L1 regularization constraints and introduces the Bayesian information criterion to suppress overfitting, thereby obtaining a sparse nonlinear dynamic expression. The third modeling subunit is used to perform parameter optimization based on the nonlinear dynamic expression, determine the sparsity intensity through cross-validation, and estimate the stability coefficient using ridge regression to obtain a set of nonlinear state equations for generating community evolution trajectories under different perturbation scenarios.

4. The comprehensive biodiversity impact assessment system based on multi-source ecological factors according to claim 3, characterized in that, The processing unit includes: The first processing subunit is used to perform spatiotemporal event construction processing on the multi-source ecological factor data. By identifying the mutation time point for each data stream, and taking the preset elevation stratification and travertine dam location mutation as a single trigger event, a spatiotemporal event sequence is obtained. The time and location accuracy of each event are corrected by a preset UAV image grid. The second processing subunit is used to perform preliminary fitting processing based on maximum likelihood estimation according to the spatiotemporal event sequence to obtain preliminary parameterized kernel coefficients and spatial background field for nonparametric kernel learning; The third processing subunit is used to perform nonparametric learning and uncertainty quantification of the kernel function based on the preliminary parameterized kernel coefficients and the spatial background field. By constructing a spacetime basis function and introducing time monotonicity and nonnegativity constraints, a spacetime intensity field that characterizes the disturbance triggering effect is obtained.

Images