Vegetation disturbance dynamic detection method and system based on remote sensing long time sequence

Abstract

The invention relates to the technical field of geological mining, in particular to a vegetation disturbance dynamic detection system and method based on a remote sensing long time sequence, and the method comprises the following steps: taking mining area surface subsidence data, mining area drought indexes and mining area meteorological data outputted by a surface subsidence mathematical model as vegetation disturbance characteristics; establishing a statistical relationship model of the surface subsidence to the vegetation biomass by utilizing a machine learning model; based on a statistical relation model of surface subsidence to vegetation biomass, a disturbance threshold value of surface subsidence to vegetation growth is quantified, and a space-time window causing maximum ecological disturbance is identified. Through long-time-sequence continuous monitoring of satellite remote sensing, the spatial-temporal variation rule among surface deformation, soil moisture and vegetation biomass is disclosed, a vegetation biomass dynamic prediction model is constructed, and real-time monitoring and prediction of crop biomass in the underground mining process are achieved; and a series of environmental effects generated by dynamic disturbance of underground mining are analyzed.

Description

Technical Field

[0001] The present invention relates to the field of geological mining technology, and in particular to a vegetation disturbance dynamic detection system based on remote sensing long time series. Background Art

[0002] Through summarizing and analyzing research on surface elevation, soil moisture, and vegetation growth in mining areas, it has been found that underground mining causes environmental changes in mining areas. With growing awareness of environmental protection, research on ecological and environmental monitoring in mining areas is increasing both domestically and internationally. With the development of remote sensing satellites, the rise of remote sensing technology, and the advancement of monitoring methods, monitoring of farmland damage in mining areas has made significant progress, but it also presents numerous opportunities and challenges.

[0003] The impact of underground mining on the surface is a dynamic process. Existing research on surface subsidence and vegetation biomass in mining areas has mostly focused on the post-mining phase, with relatively little research on the dynamic impacts caused during the underground mining process. This research is disconnected from the underground mining process and mining cycle, and can only determine the growth of above-ground vegetation or soil moisture corresponding to a specific period of underground mining. This research lacks periodicity and integration with underground mining mechanisms. Furthermore, existing research on vegetation disturbance in mining areas has focused on a relatively limited set of factors, including vegetation growth factors, underground mining parameters, soil moisture, and surface subsidence. Summary of the Invention

[0004] The purpose of the present invention is to provide a vegetation disturbance dynamic detection system and method based on remote sensing long time series to solve the technical problems in the existing technology of single monitoring factors for vegetation disturbance and the lack of integration of vegetation disturbance monitoring with underground mining mechanisms.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A method for dynamic detection of vegetation disturbance based on remote sensing long time series includes the following steps:

[0007] The SBAS-InSAR technology was used to monitor the surface deformation data caused by underground mining in the mining area, and the probability integral method was used to establish a surface subsidence mathematical model that describes the surface movement and deformation laws caused by underground mining in the mining area;

[0008] Using remote sensing technology to sample high-resolution hyperspectral remote sensing images of the mining area, and using machine learning models to establish a biomass statistical model that describes the mapping relationship between vegetation index, texture characteristics and biomass;

[0009] Calculate the mining area drought index using high-resolution hyperspectral remote sensing images;

[0010] Monitor mining area meteorological data through meteorological equipment;

[0011] Using the mining area surface subsidence data output by the surface subsidence mathematical model, the mining area drought index, and the mining area meteorological data as vegetation disturbance characteristics, a machine learning model was used to establish a statistical relationship model between surface subsidence and vegetation biomass, which describes the mapping relationship between vegetation disturbance characteristics and the mining area biomass output by the biomass statistical model.

[0012] Based on the statistical relationship model of surface subsidence and vegetation biomass, the disturbance threshold of surface subsidence on vegetation growth is quantified, and the spatiotemporal windows that cause maximum ecological disturbance are identified.

[0013] As a preferred embodiment of the present invention, the method for constructing the surface subsidence mathematical model includes:

[0014] The surface deformation data generated by underground mining in the mining area obtained through SBAS-InSAR technology is used as the parameter inversion measured value M of the surface subsidence mathematical model. j , where the parameters of the mathematical model of surface subsidence include mining parameters and geological condition parameters;

[0015] The parameters of the surface subsidence mathematical model are inverted using the universal likelihood uncertainty estimation method GLUE, including:

[0016] Step 1: Determine the physical value range of the parameters of the surface subsidence mathematical model, assuming that the parameters obey a uniform distribution within the value range;

[0017] Step 2: Using Monte Carlo and other methods, randomly generate a large number of parameter groups Y within the physical value range of the surface subsidence mathematical model parameters;

[0018] Step 3: Substitute each set of parameters into the surface subsidence mathematical model and simulate and output the surface subsidence data E j , and use the Nash coefficient NSE as the likelihood function to measure E j With M j The degree of fit is:

[0019]

[0020] Where, L(θ i |Y) is used to measure E j With M j The Nash coefficient NSE of the degree of fitting, Y is the parameter value of the parameter group, θ i is the likelihood value of the i-th group of parameters, E j Substitute the i-th group of parameters into the surface subsidence mathematical model to simulate the output of the surface subsidence data, M j represents the measured value corresponding to the i-th group of parameters, is the average value of N groups of measured values, where N is the total number of measured values. The closer NSE is to 1, the better the simulation results and the higher the likelihood value of the parameter.

[0021] Step 4: Based on the Bayesian formula, combine the parameter B in the parameter group Y i The prior distribution P(B i ) and the likelihood value P(A|B i ), calculate the parameter B i The posterior distribution of the parameters is obtained, and the uncertainty of the surface subsidence mathematical model parameters and the uncertainty range of the surface subsidence mathematical model estimation under a specific confidence interval are analyzed based on the posterior distribution results of the parameters. i The posterior distribution of is:

[0022]

[0023] In the formula, P(B i |A) is parameter B i The posterior distribution of P(B i ) is parameter B i The prior distribution of P(A|B i ) is parameter B i The likelihood value, P(B k ) is parameter B k The prior distribution of P(A|B k ) is parameter B k The likelihood value of , n is the total number of parameters in the parameter group Y;

[0024] Step 5: Select the parameter group corresponding to the high likelihood value in the posterior distribution as the parameters of the surface subsidence mathematical model to construct the surface subsidence mathematical model.

[0025] As a preferred embodiment of the present invention, the method for constructing the biomass statistical model includes:

[0026] Select M monitoring time points and obtain high-resolution hyperspectral remote sensing images of the mining area at each time point in turn;

[0027] In each high-resolution hyperspectral remote sensing image, the vegetation index is calculated based on the reflectance of the remote sensing band:

[0028] VCI i =(NDVI i -NDVI min ) / (NDVI max -NDVI min );

[0029] NDVI i =(R Redge,i -R Red,i ) / (R Redge,i +R Red,i );

[0030] Where, VCI i is the vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI i is the normalized vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Redge,i is the reflectance of the red edge band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Red,i is the reflectance of the red band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI max and NDVI min All NDVIs are i The maximum and minimum values ​​in ;

[0031] The gray-level co-occurrence matrix method (GLCM) is used to extract texture features from each high-resolution hyperspectral remote sensing image.

[0032] Through field sampling or existing databases, the measured values ​​of aboveground biomass synchronized with high-resolution hyperspectral remote sensing images are collected to obtain the biomass at M monitoring time points;

[0033] The vegetation index and texture features at M monitoring time points and the biomass at M monitoring time points are combined into a dataset, and the dataset is divided into a training set and a test set;

[0034] In the training set, the machine learning model is trained to obtain the biomass statistical model:

[0035] K4=Model1(C1,C2);

[0036] Where K4 is the biomass identifier, C1 is the vegetation index identifier, C2 is the texture feature identifier, and Model1 is the identifier of the machine learning model;

[0037] In the test set, the performance of the biomass statistical model was evaluated based on the evaluation indicators.

[0038] As a preferred solution of the present invention, the drought index is quantified using any one of the vertical drought index PDI, the improved vertical drought index MPDI, the soil moisture monitoring model SMMRS, and the contrast drought monitoring index RDMI.

[0039] As a preferred embodiment of the present invention, the method for constructing a statistical relationship model of surface subsidence and vegetation biomass includes:

[0040] m mining areas at T time points are selected, and the surface subsidence data of each mining area at each time point is outputted by the surface subsidence mathematical model in sequence, the drought index of each mining area at each time point is calculated in sequence, the meteorological data of each mining area at each time point is obtained in sequence, and the biomass of each mining area at each time point is outputted by the biomass statistical model in sequence;

[0041] Quantify the change rate of surface subsidence data, drought index and meteorological data at each mining area at each time point, and map the change rate to the weight of surface subsidence data, drought index and meteorological data in the vegetation disturbance characteristics. Then, use the weighted surface subsidence data, drought index and meteorological data to obtain the indicator disturbance characteristics of each mining area at each time point.

[0042] The disturbance index characteristics of each mining area at each time point and the biomass of each mining area at each time point are combined into a data set and divided into a training set and a test set;

[0043] In the training set, the machine learning model is trained to obtain a statistical relationship model of the surface subsidence and vegetation biomass;

[0044] In the test set, the performance of the statistical relationship model of surface subsidence on vegetation biomass was evaluated based on the evaluation indicators.

[0045] As a preferred embodiment of the present invention, a method for quantifying the disturbance threshold of ground subsidence on vegetation growth includes:

[0046] Obtain the natural fluctuation range and mean range of biomass in each mining area;

[0047] By w1 s,t The dynamic disturbance threshold of surface subsidence on vegetation growth is constructed as follows:

[0048]

[0049] Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,t is the surface subsidence data of the s-th mining area at the t-th time point.

[0050] As a preferred embodiment of the present invention, the method for identifying the spatiotemporal window that causes the maximum ecological disturbance includes:

[0051] The biomass data at each time point in each mining area output by the statistical relationship model of surface subsidence on vegetation biomass were simultaneously subjected to sliding window method and K-means spatial cluster analysis to generate a time cumulative effect map and a subsidence hotspot map respectively;

[0052] Use the space-time cube to conduct space-time interactive analysis on the time cumulative effect map and the subsidence hotspot map to count the space-time hotspot areas and obtain the space-time window map;

[0053] The total weight of each window unit in the spatiotemporal window graph is calculated, and the total weight is mapped to the spatiotemporal window graph to obtain the spatiotemporal weight heat map. The weight of the window unit is:

[0054]

[0055] Where w s,t is the weight of the window unit at the (s, t) position, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 s,t K2 s,t The weight of w3 s,t K3 s,t The weight of the window unit in the spatiotemporal window graph is min(w), and max(w) is the maximum value of the window unit weight in the spatiotemporal window graph.

[0056] In the spatiotemporal weight heat map, the area exceeding the preset conditions is marked as the disturbance peak area;

[0057] The space-time window area corresponding to the disturbance peak area in the space-time window diagram is taken as the space-time window that causes the maximum ecological disturbance.

[0058] As a preferred embodiment of the present invention, the present invention provides a vegetation disturbance dynamic detection system based on remote sensing long time series, which is applied to a vegetation disturbance dynamic detection method based on remote sensing long time series. The system includes:

[0059] The data acquisition unit is used to monitor surface deformation data generated by underground mining in the mining area using SBAS-InSAR technology, and to establish a surface subsidence mathematical model describing the surface movement and deformation caused by underground mining using the probability integral method; to sample high-resolution hyperspectral remote sensing images of the mining area using remote sensing technology, and to establish a biomass statistical model describing the mapping relationship between vegetation index, texture characteristics, and biomass using a machine learning model; and to monitor meteorological data in the mining area using meteorological equipment;

[0060] a data processing unit for calculating a mining area drought index using high-resolution hyperspectral remote sensing images; using mining area surface subsidence data output by a surface subsidence mathematical model, the mining area drought index, and mining area meteorological data as vegetation disturbance characteristics, and using a machine learning model to establish a statistical relationship model of surface subsidence to vegetation biomass that describes a mapping relationship between the vegetation disturbance characteristics and the mining area biomass output by a biomass statistical model;

[0061] The disturbance analysis unit is used to quantify the disturbance threshold of surface subsidence on vegetation growth based on the statistical relationship model of surface subsidence on vegetation biomass, and to identify the spatiotemporal windows that cause the maximum ecological disturbance.

[0062] As a preferred solution of the present invention, the statistical relationship model of surface subsidence to vegetation biomass constructed by the data processing unit is:

[0063] K4 s,t =Model2(w1 s,t *K1 s,t ,w2 s,t *K2 s,t ,w3 s,t *K3 s,t );

[0064] Where K4 s,t is the biomass of the sth mining area at the tth time point, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 st K2 st The weight of w3 st K3 st The weight of Model2 is the identifier of the machine learning model;

[0065]

[0066] Where K1 s,t-1 is the surface subsidence data of the sth mining area at the t-1th time point, K2 s,t-1 is the drought index of the sth mining area at the t-1th time point, K3 s,t-1 is the meteorological data of the sth mining area at the t-1th time point, and γ is the sensitivity adjustment coefficient.

[0067] As a preferred solution of the present invention, the dynamic disturbance threshold obtained by analysis in the disturbance analysis unit is:

[0068]

[0069] Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,t is the surface subsidence data of the s-th mining area at the t-th time point.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The present invention reveals the spatiotemporal variation patterns among surface deformation, soil moisture, and vegetation biomass through long-term continuous monitoring using satellite remote sensing. It constructs a dynamic prediction model for vegetation biomass that characterizes the complex coupling relationship among surface deformation, soil moisture, and vegetation biomass at different mining stages. This enables real-time monitoring and prediction of crop biomass during underground mining, and analyzes a series of environmental effects caused by the dynamic disturbance of underground mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0073] Figure 1 A flow chart of a method for dynamic detection of vegetation disturbance based on long-term remote sensing time series provided by an embodiment of the present invention;

[0074] Figure 2 A block diagram of a vegetation disturbance dynamic detection system based on remote sensing long time series provided by an embodiment of the present invention;

[0075] Figure 3 A schematic diagram of the RDMI model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0077] like Figure 1 As shown, the present invention provides a method for dynamic detection of vegetation disturbance based on remote sensing long time series, comprising the following steps:

[0078] The SBAS-InSAR technology was used to monitor the surface deformation data caused by underground mining in the mining area, and the probability integral method was used to establish a surface subsidence mathematical model that describes the surface movement and deformation laws caused by underground mining in the mining area;

[0079] Using remote sensing technology to sample high-resolution hyperspectral remote sensing images of the mining area, and using machine learning models to establish a biomass statistical model that describes the mapping relationship between vegetation index, texture characteristics and biomass;

[0080] Calculate the mining area drought index using high-resolution hyperspectral remote sensing images;

[0081] Monitor mining area meteorological data through meteorological equipment;

[0082] Using the mining area surface subsidence data output by the surface subsidence mathematical model, the mining area drought index, and the mining area meteorological data as vegetation disturbance characteristics, a machine learning model was used to establish a statistical relationship model between surface subsidence and vegetation biomass, which describes the mapping relationship between vegetation disturbance characteristics and the mining area biomass output by the biomass statistical model.

[0083] Based on the statistical relationship model of surface subsidence and vegetation biomass, the disturbance threshold of surface subsidence on vegetation growth is quantified, and the spatiotemporal windows that cause maximum ecological disturbance are identified.

[0084] The present invention aims to realize dynamic monitoring of vegetation disturbance caused by underground mining in mining areas. It establishes a complex coupling relationship between surface deformation, soil moisture and vegetation biomass at different mining stages, analyzes the series of environmental effects caused by the dynamic disturbance of underground mining, and reveals the temporal and spatial variation patterns among surface deformation, soil moisture and vegetation biomass.

[0085] In order to establish the complex coupling relationship between surface deformation, soil moisture and vegetation biomass at different mining stages, the present invention uses a machine learning model to map and associate vegetation disturbance characteristics composed of mining area surface subsidence data, mining area drought index, and mining area meteorological data with vegetation biomass, thereby constructing a statistical relationship model of surface subsidence to vegetation biomass corresponding to the complex coupling relationship between surface deformation, soil moisture and vegetation biomass at different mining stages, and realizing real-time monitoring and prediction of crop biomass based on mining area surface subsidence data, mining area drought index, and mining area meteorological data during underground mining.

[0086] Among them, the present invention adopts a mathematical model for subsidence prediction based on the probability integral method to obtain surface subsidence data, takes the surface subsidence data obtained by SBAS-InSAR as the measured value, uses the GLUE algorithm to perform parameter inversion, and realizes the prediction of subsidence for underground mining. The present invention adopts the General Likelihood Uncertainty Estimation (GLUE) method for parameter estimation. The main advantage of GLUE based on the Bayesian principle compared to the deterministic inversion algorithm is that it can provide the posterior distribution of the inversion parameters, so that the uncertainty of the model parameters and the uncertainty range of the model estimation under a certain confidence interval can be analyzed, so that the present invention obtains accurate prediction of surface subsidence data, which helps to ensure the model performance of the subsequent statistical relationship model of surface subsidence to vegetation biomass.

[0087] The methods for constructing mathematical models of surface subsidence include:

[0088] The surface deformation data generated by underground mining in the mining area obtained through SBAS-InSAR technology is used as the parameter inversion measured value M of the surface subsidence mathematical model. j , where the parameters of the mathematical model of surface subsidence include mining parameters (such as mining thickness, subsidence coefficient, main influencing angle, etc.) and geological condition parameters (such as rock stratum mechanical properties);

[0089] The parameters of the surface subsidence mathematical model are inverted using the universal likelihood uncertainty estimation method GLUE, including:

[0090] Step 1: Determine the physical value range of the surface subsidence mathematical model parameters (such as the subsidence coefficient and horizontal movement coefficient), assuming that the parameters follow a uniform distribution within the value range (without additional prior information);

[0091] Step 2: Using Monte Carlo and other methods, randomly generate a large number of parameter groups Y within the physical value range of the surface subsidence mathematical model parameters;

[0092] Step 3: Substitute each set of parameters into the surface subsidence mathematical model and simulate and output the surface subsidence data Ej , and use the Nash coefficient NSE as the likelihood function to measure E j With M j The degree of fit is:

[0093]

[0094] Where, L(θ i |Y) is used to measure E j With M j The Nash coefficient NSE of the degree of fitting, Y is the parameter value of the parameter group, θ i is the likelihood value of the i-th group of parameters, E j Substitute the i-th group of parameters into the surface subsidence mathematical model to simulate the output of the surface subsidence data, M j represents the measured value corresponding to the i-th group of parameters, is the average value of N groups of measured values, where N is the total number of measured values. The closer NSE is to 1, the better the simulation results and the higher the likelihood value of the parameter.

[0095] Step 4: Based on the Bayesian formula, combine the parameter B in the parameter group Y i The prior distribution P(B i ) and the likelihood value P(A|B i ), calculate the parameter B i The posterior distribution of the parameters is obtained, and the uncertainty of the parameters of the surface subsidence mathematical model and the uncertainty range of the surface subsidence mathematical model estimation under a specific confidence interval are analyzed based on the posterior distribution of the parameters. i The posterior distribution of is:

[0096]

[0097] In the formula, P(B i |A) is parameter B i The posterior distribution of P(B i ) is parameter B i The prior distribution of P(A|B i ) is parameter B i The likelihood value, P(B k ) is parameter B k The prior distribution of P(A|B k ) is parameter B k The likelihood value of , n is the total number of parameters in the parameter group Y;

[0098] Step 5: Select the parameter group corresponding to the high likelihood value in the posterior distribution as the parameters of the surface subsidence mathematical model to construct the surface subsidence mathematical model.

[0099] The acquisition of vegetation biomass in the present invention is also automatically acquired using a machine learning algorithm. First, satellite remote sensing data (high-resolution hyperspectral remote sensing images) of different time periods are obtained through remote sensing technology, and the vegetation coverage in the mining area is extracted. The vegetation index and texture characteristics related to biomass are analyzed in the satellite remote sensing data, and a biomass statistical model is constructed with the above-ground biomass as the dependent variable and the vegetation index and texture characteristics related to biomass as the independent variables to achieve automatic identification of vegetation. There is no need to carry out vegetation biomass sampling processes such as "collecting soil samples in the mining area through field sampling, including soils at different depths and under different vegetation coverage, to obtain key parameters such as soil physical and chemical properties and moisture content. In addition, a vegetation survey will be conducted to record information such as vegetation type, coverage, and biomass", thereby improving the efficiency of obtaining biomass data.

[0100] The methods for constructing biomass statistical models include:

[0101] Select M monitoring time points and obtain high-resolution hyperspectral remote sensing images of the mining area at each time point in turn;

[0102] In each high-resolution hyperspectral remote sensing image, the vegetation index is calculated based on the reflectance of the remote sensing band:

[0103] VCI i =(NDVI i -NDVI min ) / (NDVI max -NDVI min );

[0104] NDVI i =(R Redge,i -R Red,i ) / (R Redge,i +R Red,i );

[0105] Where, VCI i is the vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI i is the normalized vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Redge,i is the reflectance of the red edge band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Red,i is the reflectance of the red band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI max and NDVI min All NDVIs are i The maximum and minimum values ​​in ;

[0106] The gray-level co-occurrence matrix method (GLCM) is used to extract texture features from each high-resolution hyperspectral remote sensing image.

[0107] Through field sampling or existing databases, the measured values ​​of aboveground biomass (dry weight or fresh weight) synchronized with high-resolution hyperspectral remote sensing images are collected to obtain the biomass at M monitoring time points;

[0108] The vegetation index and texture features at M monitoring time points and the biomass at M monitoring time points are combined into a dataset, and the dataset is divided into a training set and a test set;

[0109] In the training set, machine learning models such as random forest (RF) and support vector machine (SVM) are trained to obtain the biomass statistical model:

[0110] K4=Model1(C1,C2);

[0111] Where K4 is the biomass identifier, C1 is the vegetation index identifier, C2 is the texture feature identifier, and Model1 is the identifier of the machine learning model;

[0112] In the test set, based on evaluation metrics such as R 2 , root mean square error RMSE, and mean absolute error MAE were used to evaluate the performance of the biomass statistical model.

[0113] The drought index is quantified using any one of the vertical drought index (PDI), improved vertical drought index (MPDI), soil moisture monitoring model (SMMRS), and contrast drought monitoring index (RDMI).

[0114] The present invention compares the reliability of different drought indices for characterizing the spatial heterogeneity of soil moisture in mining areas, and selects or constructs a drought index with higher accuracy based on the characteristics of the mining areas, where:

[0115] The Perpendicular Drought Index (PDI):

[0116]

[0117] Where R Redge and R Red are the reflectances of the red edge band and the red band, respectively, and M is the slope of the soil line.

[0118] The Perpendicular Drought Index (PDI):

[0119]

[0120] Where R Redge and R Red are the reflectances of the red edge band and the red band, respectively.

[0121]

[0122] R l =f v R v,l +(1-f v )R s,l ;

[0123]

[0124] Where R s,l is the soil reflectance value in band l, R l is the reflectivity value in band l, R v,l is the vegetation reflectance value in band l, R v,l Can be considered as a coefficient of known vegetation growth.

[0125] Soil Moisture Monitoring of Remote Sensing (SMMRS):

[0126]

[0127] Where R Redge and R Red are the reflectances of the red edge band and the red band, respectively, and M is the slope of the soil line.

[0128] The proposed Ratio Dryness Monitoring Index (RDMI) is constructed based on the principle of Figure 3 , line AB is the edge of the soil, and line AC represents the minimum water stress area under various vegetation cover conditions. Line AC and line BC are defined as the wet edge and dry edge of the Red-Redge spectral feature space, respectively. The position of a random pixel in the approximate triangle formed by the three lines AB, AC, and BC contains two kinds of information. The first is the vegetation cover condition, which is expressed as the distance from a given pixel to the soil edge (AB). The second is the water stress state under the same vegetation cover conditions, which can be expressed as the ratio of the distance from a given pixel to the wet edge (AC) to the distance from the wet edge to the dry edge (BC).

[0129] R Redge =M wet R Red +I wet ;

[0130] Where M wet and Iwet are the slope and intercept of the wet side, respectively.

[0131] The calculation formula for dry edge (BC) is as follows:

[0132] R Redge =M dry R Red +I dry ;

[0133] Where M dry and I dry represent the slope and intercept of the stem side, respectively.

[0134] The methods for constructing the statistical relationship model of surface subsidence and vegetation biomass include:

[0135] m mining areas at T time points are selected, and the surface subsidence data of each mining area at each time point is outputted by the surface subsidence mathematical model in sequence, the drought index of each mining area at each time point is calculated in sequence, the meteorological data of each mining area at each time point is obtained in sequence, and the biomass of each mining area at each time point is outputted by the biomass statistical model in sequence;

[0136] Quantify the change rate of surface subsidence data, drought index and meteorological data at each mining area at each time point, and map the change rate to the weight of surface subsidence data, drought index and meteorological data in the vegetation disturbance characteristics. Then, use the weighted surface subsidence data, drought index and meteorological data to obtain the indicator disturbance characteristics of each mining area at each time point.

[0137] The disturbance index characteristics of each mining area at each time point and the biomass of each mining area at each time point are combined into a data set and divided into a training set and a test set;

[0138] In the training set, the machine learning model is trained to obtain a statistical relationship model between surface subsidence and vegetation biomass;

[0139] In the test set, the performance of the statistical relationship model of surface subsidence on vegetation biomass was evaluated based on the evaluation indicators.

[0140] The statistical relationship model of surface subsidence to vegetation biomass in the present invention is:

[0141] K4 s,t =Model2(w1 s,t *K1 s,t ,w2 s,t *K2 s,t ,w3 s,t *K3 s,t );

[0142] Where K4 s,tis the biomass of the sth mining area at the tth time point, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 st K2 st The weight of w3 st K3 st The weight of Model2 is the identifier of the machine learning model;

[0143]

[0144] Where K1 s,t-1 is the surface subsidence data of the sth mining area at the t-1th time point, K2 s,t-1 is the drought index of the sth mining area at the t-1th time point, K3 s,t-1 is the meteorological data of the sth mining area at the t-1th time point, and γ is the sensitivity adjustment coefficient.

[0145] When constructing a statistical relationship model between surface subsidence and vegetation biomass, the present invention performs weighted processing on each feature of the indicator disturbance feature. The weight is related to the change rate of surface subsidence data, drought index, and meteorological data, and the weight is positively correlated with the degree of change of each variable (the greater the change, the higher the weight). Its advantages are mainly reflected in the following aspects:

[0146] First, enhance the model's sensitivity to key driving factors and dynamically capture important periods of vegetation disturbance: if a variable changes dramatically in a specific period or region (such as sudden subsidence, extreme drought), a high weight will amplify the impact of the variable on biomass, making the model more sensitive to short-term or local ecological disturbances. Example: The amount of subsidence in the rainy season increases sharply due to fluctuations in groundwater levels. The model quickly identifies its destructive effects on vegetation through high weights. Distinguishable dominant factors: In a complex system where multiple variables coexist, dynamic weights can automatically screen out the core factors that dominate biomass changes in the current environment. For example, in the dry season, the degree of change in the drought index may be higher, and its increased weight can more accurately quantify the impact of water stress.

[0147] Second, improve the spatiotemporal adaptability of the model and model spatiotemporal heterogeneity: the subsidence rate and degree of drought in different areas of the mining area may vary significantly. Through dynamic weights, the model can adaptively adjust parameters for different sub-areas or time periods to avoid the "one-size-fits-all" static assumption. For example: in the core area of ​​the mining area where subsidence is active, the weight of subsidence is automatically increased; in the marginal area, the weight of meteorological factors may dominate. Responsive to hysteresis effects: If there is a time lag in the impact of subsidence on biomass (such as vegetation begins to decline several months after subsidence), dynamic weights can capture the changing trend of the lag period through sliding window analysis and optimize the temporal resolution of the model.

[0148] Third, it improves the ability to predict extreme events. High weighting of extreme events: When extreme weather (such as persistent high temperatures) or sudden geological disasters (such as landslides) occur, the degree of change in related variables increases dramatically. High weighting allows the model to pay more attention to these outliers, thereby improving the accuracy of predicting extreme ecological disturbances. For example, if a sudden mining operation causes a surge in subsidence, the model can quickly identify its immediate impact on biomass through dynamic weighting. It can also reduce noise interference during stable periods: During periods of slow variable change (such as stable mining), low weighting can suppress the interference of stable noise on the model and enhance overall robustness.

[0149] Fourth, it supports dynamic threshold and spatiotemporal window identification, as well as dynamic calibration of disturbance thresholds. By linking weights with the degree of change, the model can dynamically adjust disturbance thresholds under different conditions. For example, when the rate of change in subsidence exceeds twice the historical average, its weight is increased, triggering sensitivity adjustments in the threshold determination logic. Precise localization of spatiotemporal windows: Highly weighted regions and time periods can be directly mapped to spatiotemporal windows with the most significant ecological disturbances. For example, by combining peak subsidence weights with fluctuations in the drought index, ecologically vulnerable periods characterized by "accelerated subsidence and overlapping droughts" can be identified.

[0150] Methods for quantifying the threshold of disturbance to vegetation growth caused by land subsidence include:

[0151] Obtain the natural fluctuation range and mean range of biomass in each mining area;

[0152] By w1 st The dynamic disturbance threshold of surface subsidence on vegetation growth is constructed as follows:

[0153]

[0154] Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,tis the surface subsidence data of the s-th mining area at the t-th time point.

[0155] Among them, dynamic calibration of disturbance threshold: through the linkage between weight and degree of change, the model can dynamically adjust the disturbance threshold under different conditions. When the variable changes drastically (w1 s,t Increase), in the formula The threshold value decreases as a whole. At this time, a small variable fluctuation can trigger the judgment, which increases the sensitivity of the model to sudden disturbances. When the variable changes slowly (w1 s,t smaller), Increase, the threshold is raised to avoid misjudging natural fluctuations as disturbance events.

[0156] The threshold sensitivity adjustment example is as follows: When the rate of change of the subsidence (more than 2 standard deviations above the historical mean), γ=0.5, w1 s,t =1+0.5*2=2, at this time, the influence of the subsidence amount is doubled, the traditional static threshold: D s =μ s +2σ s , assuming μ s =100mm,σ s =20mm, D s =100+20*2=140mm, the present invention When the sudden increase in the amount of subsidence causes the weight w1 s,t =2, the threshold is reduced from 140 mm to 120 mm. The model is more sensitive to subsidence and can more easily detect ecological disturbances. Therefore, it can respond quickly to extreme events (such as sudden mining) to avoid missed detections, and increase the threshold during stable periods to suppress noise interference.

[0157] Methods for identifying spatiotemporal windows that trigger maximum ecological disturbances include:

[0158] The biomass data at each time point in each mining area output by the statistical relationship model of surface subsidence on vegetation biomass were simultaneously subjected to sliding window method and K-means spatial cluster analysis to generate a time cumulative effect map and a subsidence hotspot map respectively;

[0159] Use the space-time cube to conduct space-time interactive analysis on the time cumulative effect map and the subsidence hotspot map to count the space-time hotspot areas and obtain the space-time window map;

[0160] The total weight of each window unit in the spatiotemporal window graph is calculated, and the total weight is mapped to the spatiotemporal window graph to obtain the spatiotemporal weight heat map. The weight of the window unit is:

[0161]

[0162] Where w s,tis the weight of the window unit at the (s, t) position, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 s,t K2 s,t The weight of w3 s,t K3 s,t The weight of the window unit in the spatiotemporal window graph is min(w), and max(w) is the maximum value of the window unit weight in the spatiotemporal window graph.

[0163] In the spatiotemporal weight heat map, the area exceeding the preset conditions is marked as the disturbance peak area;

[0164] The space-time window area corresponding to the disturbance peak area in the space-time window diagram is taken as the space-time window that causes the maximum ecological disturbance.

[0165] This method enables precise location of spatiotemporal windows. High-weighted regions and time periods can be directly mapped to spatiotemporal windows where ecological disturbances (such as biomass declines) occur most significantly at specific times and locations. By analyzing the spatiotemporal distribution patterns of high-weighted variables (such as subsidence and drought index), "disturbance peaks" can be located.

[0166] Preset condition combination: Subsidence weight w1 s,t >0.8 (top 20% quantile); drought index weight w2 s,t >0.8; meteorological factor weight w3 s,t >0.8; when the above conditions are met at the same time, it is determined to be an "ecologically fragile window".

[0167] For example, the monthly change rate of subsidence in region A suddenly increases by 200% (w1 s,t =0.9), and the PDI index decreased by 30% during the same period (indicating the intensification of drought, w2 s,t =0.85w), meteorological factor: no precipitation for 30 consecutive days (indicating abnormal weather, w3 s,t =0.9).

[0168] Judgment result: Area A is marked as the "maximum disturbance window" in month t, and ecological restoration must be initiated as a priority.

[0169] Therefore, high-weighted regions and time periods can be directly mapped to spatiotemporal windows of greatest ecological disturbance. For example, by combining peak subsidence weights with fluctuations in the drought index, we can identify periods of ecological vulnerability characterized by "accelerated subsidence and combined drought." This directly targets high-risk spatiotemporal units and optimizes resource allocation. By decomposing these weights, we can identify the dominant factor (e.g., whether subsidence or drought is the primary driver).

[0170] In summary, by setting dynamic weights based on the degree of change for subsidence, drought index, and meteorological data, the model can accurately capture key driving factors and improve the ability to respond to extreme events; adapt to spatiotemporal heterogeneity and optimize the recognition accuracy of thresholds and windows; balance sensitivity and robustness to provide reliable ecological disturbance assessment in complex environments.

[0171] like Figure 2 As shown, the present invention provides a vegetation disturbance dynamic detection system based on remote sensing long time series, which is applied to a vegetation disturbance dynamic detection method based on remote sensing long time series. The system includes:

[0172] The data acquisition unit is used to monitor surface deformation data generated by underground mining in the mining area using SBAS-InSAR technology, and to establish a surface subsidence mathematical model describing the surface movement and deformation caused by underground mining using the probability integral method; to sample high-resolution hyperspectral remote sensing images of the mining area using remote sensing technology, and to establish a biomass statistical model describing the mapping relationship between vegetation index, texture characteristics, and biomass using a machine learning model; and to monitor meteorological data in the mining area using meteorological equipment;

[0173] a data processing unit for calculating a mining area drought index using high-resolution hyperspectral remote sensing images; using mining area surface subsidence data output by a surface subsidence mathematical model, the mining area drought index, and mining area meteorological data as vegetation disturbance characteristics, and using a machine learning model to establish a statistical relationship model of surface subsidence to vegetation biomass that describes a mapping relationship between the vegetation disturbance characteristics and the mining area biomass output by a biomass statistical model;

[0174] The disturbance analysis unit is used to quantify the disturbance threshold of surface subsidence on vegetation growth based on the statistical relationship model of surface subsidence on vegetation biomass, and to identify the spatiotemporal windows that cause the maximum ecological disturbance.

[0175] The statistical relationship model between surface subsidence and vegetation biomass constructed by the data processing unit is:

[0176] K4 s,t =Model2(w1 s,t *K1 s,t ,w2 s,t *K2 s,t ,w3 s,t *K3 s,t );

[0177] Where, K4 s,t is the biomass of the sth mining area at the tth time point, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 s,t K2 s,t The weight of w3 s,t K3 s,t The weight of Model2 is the identifier of the machine learning model;

[0178]

[0179] Where, K1 s,t-1 is the surface subsidence data of the sth mining area at the t-1th time point, K2 s,t-1 is the drought index of the sth mining area at the t-1th time point, K3 s,t-1 is the meteorological data of the sth mining area at the t-1th time point, and γ is the sensitivity adjustment coefficient.

[0180] The dynamic disturbance threshold obtained by analysis in the disturbance analysis unit is:

[0181]

[0182] Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,t is the surface subsidence data of the s-th mining area at the t-th time point.

[0183] The present invention reveals the spatiotemporal variation patterns among surface deformation, soil moisture, and vegetation biomass through long-term continuous monitoring using satellite remote sensing. It constructs a dynamic prediction model for vegetation biomass that characterizes the complex coupling relationship among surface deformation, soil moisture, and vegetation biomass at different mining stages. This enables real-time monitoring and prediction of crop biomass during underground mining, and analyzes a series of environmental effects caused by the dynamic disturbance of underground mining.

[0184] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. A method for dynamic detection of vegetation disturbance based on remote sensing long time series, characterized in that: The following steps are involved: The SBAS-InSAR technology was used to monitor the surface deformation data caused by underground mining in the mining area, and the probability integral method was used to establish a surface subsidence mathematical model that describes the surface movement and deformation laws caused by underground mining in the mining area; Using remote sensing technology to sample high-resolution hyperspectral remote sensing images of the mining area, and using machine learning models to establish a biomass statistical model that describes the mapping relationship between vegetation index, texture characteristics and biomass; Calculate the mining area drought index using high-resolution hyperspectral remote sensing images; Monitor mining area meteorological data through meteorological equipment; Using the mining area surface subsidence data output by the surface subsidence mathematical model, the mining area drought index, and the mining area meteorological data as vegetation disturbance characteristics, a machine learning model was used to establish a statistical relationship model between surface subsidence and vegetation biomass, which describes the mapping relationship between vegetation disturbance characteristics and the mining area biomass output by the biomass statistical model. Based on the statistical relationship model of surface subsidence and vegetation biomass, the disturbance threshold of surface subsidence on vegetation growth is quantified, and the spatiotemporal windows that cause maximum ecological disturbance are identified.

2. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 1, characterized in that: The method for constructing the surface subsidence mathematical model includes: The surface deformation data generated by underground mining in the mining area obtained by SBAS-InSAR technology is used as the parameter inversion measured value M of the surface subsidence mathematical model. j , where the parameters of the mathematical model of surface subsidence include mining parameters and geological condition parameters; The parameters of the surface subsidence mathematical model are inverted using the universal likelihood uncertainty estimation method GLUE, including: Step 1: Determine the physical value range of the parameters of the surface subsidence mathematical model, assuming that the parameters obey a uniform distribution within the value range; Step 2: Using Monte Carlo and other methods, randomly generate a large number of parameter groups Y within the physical value range of the surface subsidence mathematical model parameters; Step 3: Substitute each set of parameters into the surface subsidence mathematical model and simulate and output the surface subsidence data E j , and use the Nash coefficient NSE as the likelihood function to measure E j With M j The degree of fit is: Where, L(θ i │Y) is used to measure E j With M j The Nash coefficient NSE of the degree of fitting, Y is the parameter value of the parameter group, θ i is the likelihood value of the i-th group of parameters, E j Substitute the i-th group of parameters into the surface subsidence mathematical model to simulate the output of the surface subsidence data, M j represents the measured value corresponding to the i-th group of parameters, is the average value of N groups of measured values, where N is the total number of measured values; Step 4: Based on the Bayesian formula, combine the parameter B in the parameter group Y i The prior distribution P(B i ) and the likelihood value P(A│B i ), calculate the parameter B i The posterior distribution of the parameters is obtained, and the uncertainty of the surface subsidence mathematical model parameters and the uncertainty range of the surface subsidence mathematical model estimation under a specific confidence interval are analyzed based on the posterior distribution results of the parameters. i The posterior distribution of is: In the formula, P(B i │A) is parameter B i The posterior distribution of P(B i ) is parameter B i Prior distribution of P(A│B i ) is parameter B i The likelihood value, P(B k ) is parameter B k Prior distribution of P(A│B k ) is parameter B k The likelihood value of , n is the total number of parameters in the parameter group Y; Step 5: Select the parameter group corresponding to the high likelihood value in the posterior distribution as the parameters of the surface subsidence mathematical model to construct the surface subsidence mathematical model.

3. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 2, characterized in that: The method for constructing the biomass statistical model comprises: Select M monitoring time points and obtain high-resolution hyperspectral remote sensing images of the mining area at each time point in turn; In each high-resolution hyperspectral remote sensing image, the vegetation index is calculated based on the reflectance of the remote sensing band: VCI i =(NDVI i ―NDVI min ) / (NDVI max ―NDVI min ); NDVI i =(R Redge,i ―R Red,i ) / (R Redge,i +R Red,i ); Where, VCI i is the vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI i is the normalized vegetation index corresponding to the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Redge,i is the reflectance of the red edge band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, R Red,i is the reflectance of the red band in the high-resolution hyperspectral remote sensing image at the i-th monitoring time point, NDVI max and NDVI min All NDVIs are i The maximum and minimum values ​​in ; The gray-level co-occurrence matrix method (GLCM) is used to extract texture features from each high-resolution hyperspectral remote sensing image. Through field sampling or existing databases, the measured values ​​of aboveground biomass synchronized with high-resolution hyperspectral remote sensing images are collected to obtain the biomass at M monitoring time points; The vegetation index and texture features at M monitoring time points and the biomass at M monitoring time points are combined into a dataset, and the dataset is divided into a training set and a test set; In the training set, the machine learning model is trained to obtain the biomass statistical model: K4=Model1(C1,C2); Where K4 is the biomass identifier, C1 is the vegetation index identifier, C2 is the texture feature identifier, and Model1 is the identifier of the machine learning model; In the test set, the performance of the biomass statistical model was evaluated based on the evaluation indicators.

4. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 3, characterized in that: The drought index is quantified by using any one of the vertical drought index PDI, the improved vertical drought index MPDI, the soil moisture monitoring model SMMRS, and the contrast drought monitoring index RDMI.

5. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 4, characterized in that: The method for constructing the statistical relationship model of surface subsidence and vegetation biomass includes: m mining areas at T time points are selected, and the surface subsidence data of each mining area at each time point is outputted by the surface subsidence mathematical model in sequence, the drought index of each mining area at each time point is calculated in sequence, the meteorological data of each mining area at each time point is obtained in sequence, and the biomass of each mining area at each time point is outputted by the biomass statistical model in sequence; Quantify the change rate of surface subsidence data, drought index and meteorological data at each mining area at each time point, and map the change rate to the weight of surface subsidence data, drought index and meteorological data in the vegetation disturbance characteristics. Then, use the weighted surface subsidence data, drought index and meteorological data to obtain the indicator disturbance characteristics of each mining area at each time point. The disturbance index characteristics of each mining area at each time point and the biomass of each mining area at each time point are combined into a data set and divided into a training set and a test set; In the training set, the machine learning model is trained to obtain a statistical relationship model of the surface subsidence and vegetation biomass; In the test set, the performance of the statistical relationship model of surface subsidence on vegetation biomass was evaluated based on the evaluation indicators.

6. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 5, characterized in that: Methods for quantifying the threshold of disturbance to vegetation growth caused by land subsidence include: Obtain the natural fluctuation range and mean range of biomass in each mining area; By w1 s,t The dynamic disturbance threshold of surface subsidence on vegetation growth is constructed as follows: Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,t is the surface subsidence data of the s-th mining area at the t-th time point.

7. The method for dynamic detection of vegetation disturbance based on remote sensing long time series according to claim 6, characterized in that: Methods for identifying spatiotemporal windows that trigger maximum ecological disturbances include: The biomass data at each time point in each mining area output by the statistical relationship model of surface subsidence on vegetation biomass were simultaneously subjected to sliding window method and K-means spatial cluster analysis to generate a time cumulative effect map and a subsidence hotspot map respectively; Use the space-time cube to conduct space-time interactive analysis on the time cumulative effect map and the subsidence hotspot map to count the space-time hotspot areas and obtain the space-time window map; The total weight of each window unit in the spatiotemporal window graph is calculated, and the total weight is mapped to the spatiotemporal window graph to obtain the spatiotemporal weight heat map. The weight of the window unit is: Where w s,t is the weight of the window unit at the (s, t) position, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 s,t K2 s,t The weight of w3 s,t K3 s,t The weight of the window unit in the spatiotemporal window graph is min(w), and max(w) is the maximum value of the window unit weight in the spatiotemporal window graph. In the spatiotemporal weight heat map, the area exceeding the preset conditions is marked as the disturbance peak area; The space-time window area corresponding to the disturbance peak area in the space-time window diagram is taken as the space-time window that causes the maximum ecological disturbance.

8. A vegetation disturbance dynamic detection system based on remote sensing long time series, characterized by: A method for dynamic detection of vegetation disturbance based on remote sensing long time series as described in any one of claims 1 to 7, the system comprising: The data acquisition unit is used to monitor surface deformation data generated by underground mining in the mining area using SBAS-InSAR technology, and to establish a surface subsidence mathematical model describing the surface movement and deformation caused by underground mining using the probability integral method; to sample high-resolution hyperspectral remote sensing images of the mining area using remote sensing technology, and to establish a biomass statistical model describing the mapping relationship between vegetation index, texture characteristics, and biomass using a machine learning model; and to monitor meteorological data in the mining area using meteorological equipment; a data processing unit for calculating a mining area drought index using high-resolution hyperspectral remote sensing images; using mining area surface subsidence data output by a surface subsidence mathematical model, the mining area drought index, and mining area meteorological data as vegetation disturbance characteristics, and using a machine learning model to establish a statistical relationship model of surface subsidence to vegetation biomass that describes a mapping relationship between the vegetation disturbance characteristics and the mining area biomass output by a biomass statistical model; The disturbance analysis unit is used to quantify the disturbance threshold of surface subsidence on vegetation growth based on the statistical relationship model of surface subsidence on vegetation biomass, and to identify the spatiotemporal windows that cause the maximum ecological disturbance.

9. The vegetation disturbance dynamic detection system based on remote sensing long time series according to claim 8, characterized in that: The statistical relationship model of surface subsidence to vegetation biomass constructed by the data processing unit is: K4 s,t =Model2(w1 s,t *K1 s,t ,w2 s,t *K2 s,t ,w3 s,t *K3 s,t ); Where, K4 s,t is the biomass of the sth mining area at the tth time point, K1 s,t is the surface subsidence data of the sth mining area at the tth time point, K2 s,t is the drought index of the sth mining area at the tth time point, K3 s,t is the meteorological data of the sth mining area at the tth time point, w1 s,t K1 s,t The weight, w2 s,t K2 s,t The weight of w3 s,t K3 s,t The weight of Model2 is the identifier of the machine learning model; Where, K1 s,t-1 is the surface subsidence data of the sth mining area at the t-1th time point, K2 s,t-1 is the drought index of the sth mining area at the t-1th time point, K3 s,t-1 is the meteorological data of the sth mining area at the t-1th time point, and γ is the sensitivity adjustment coefficient.

10. The vegetation disturbance dynamic detection system based on remote sensing long time series according to claim 9, characterized in that: The dynamic disturbance threshold obtained by analysis in the disturbance analysis unit is: Where D s,t is the disturbance threshold of the sth mining area at the tth time point, μ s is the mean range of biomass in the sth mining area, σ s is the natural fluctuation range of the sth mining area, w1 s,t K1 s,t The weight, K1 s,t is the surface subsidence data of the s-th mining area at the t-th time point.

Images